Posts

What Is an AI Citation Audit & What Can It Tell You About Your Content

/0 Comments/in /by

Key Takeaways

  • An AI citation audit tells you, on a per-topic and per-platform basis, where your visibility gaps come from and what type of action closes each one.
  • The majority of citations driving AI responses typically come from third-party sources, not brand-owned pages. Competitors appear because independent sites reference them, not because their own content is being surfaced.
  • High-volume, low-differentiation content faces the highest displacement risk in an AI environment. Generic how-to guides are exactly the type of content AI can synthesize without sending users anywhere.
  • The goal of content strategy shifts from answering every possible question to being present with genuine authority in the specific contexts that matter to your buyers.

If you’ve been tracking your brand in AI tools and wondering why the data isn’t telling you anything useful, the problem is usually upstream: generic prompts, the wrong measurement model, inputs that don’t reflect how real buyers actually search. In an earlier piece, I introduced a structured framework for fixing it. This post is about what happens once the framework does its job.

Once you have well-constructed prompts, two layers of metrics, and a clear picture of where your brand appears across AI platforms, you get a specific and actionable output: a citation audit. Understanding what is an AI audit and what it tells you is where measurement becomes strategy.

The citation audit sorts your visibility gaps into three categories: gaps that require digital PR, gaps that require owned content, and gaps that point to social and community management. Each category demands a different type of response. And the pattern running across all of them points to the same conclusion: the content playbook built around maximizing coverage and keyword volume is losing ground to one built around genuine authority and relevance.

This post makes that argument concrete, and closes the argument with the strategic implication that follows.

What the Citation Audit Actually Shows

Once the structured topical analysis is complete, the methodology exports citation data for the highest-opportunity topics on each platform. That data breaks down across three dimensions.

Third-party content accounts for the bulk of what AI is drawing on. In most audits, well over 80 percent of highly cited pages come from independent sources: sector publications, accounting and advisory firm blogs, business setup consultancies, and regulatory guides. These are not the brand’s own pages. They are pages where the brand (or a competitor) is mentioned in the context of explaining something broader.

Owned content plays a smaller role than most teams expect, but it’s not irrelevant. Specific owned pages, particularly long-form guides that cover a topic with genuine depth, do earn citations. The issue is that most brands’ owned content skews toward service pages and thin category coverage, which AI systems have little reason to cite when better third-party resources exist.

Social and UGC signals are a smaller but growing dimension. Platforms like Reddit and Quora appear in citation data for certain topic types, particularly those involving peer experience, comparisons, and community knowledge. This is an underserved channel for most brands.

The example below shows how this ecosystem applied to one NP Digital client that we worked with.

The executive summary of results from an AI visibility audit NP Digital conducted for a client.

In one audit, roughly 80 percent of highly cited pages for compliance-related topics came from independent accounting, tax, and audit firms. The brand’s own content was rarely surfaced. Competitors appeared not because of anything they had published directly, but because third-party sites were using them as examples when explaining regulations and requirements. Visibility was earned indirectly, through the content ecosystem, not through the brand’s own pages.

The Coverage Trap

To understand why this matters strategically, it helps to understand the model it’s replacing.

The coverage mindset that drove SEO content strategy for the past decade wasn’t irrational. Traffic was the primary currency. Search engines rewarded breadth. The more questions you could answer, the more pages you could rank, and the more traffic you could capture and convert at the margin. Publishing at volume made sense.

Alt text: Two-column diagram contrasting devalued generic content types on the left with high-value authoritative content types on the right, illustrating the shift from coverage to authority in an AI search environment.]

That model is breaking down in an AI environment, and the citation audit is where you see it most clearly.

AI systems are built to synthesize and summarize. Content that exists to answer broad, generic questions is exactly the type of content AI can handle on its own, without sending users anywhere. A page explaining what SEO is, or listing the top ten CRM tools, or walking through a basic how-to process is precisely the type of content that gets absorbed into an AI response rather than cited as a source.

The more your content resembles what an AI would generate from a basic prompt, the less reason an AI has to cite you. This is the coverage trap: scaling the old model doesn’t just fail to improve AI visibility; it actively increases exposure to displacement.

A graphic showing content strategy is shifting from SEO content to original research and authority.

What AI Systems Actually Cite

The citation audit goes beyond revealing gaps to reveal patterns in what earns citations, and that pattern is consistent across topics and platforms.

Citations go to content that demonstrates genuine expertise in a specific context versus the biggest brand or highest-traffic page. Original research with proprietary data. Long-form guides that go deeper than the obvious. First-hand experience presented with authority. Comparison content that places competitors in context rather than avoiding them.

The pattern from real audit work: educational long-form guides consistently outperform service pages. Content that mentions competitors as examples within broader category coverage drives more citations than content focused exclusively on the brand. Pages that answer a specific, high-intent question with real depth earn citations.

This is a function of what the content actually contains. AI systems are drawing on content that has established a genuine association with a concept, problem, or use case. That association is built through depth, specificity, and demonstrable expertise, not through breadth of coverage.

Table showing that for both compliance and banking topics, long-form educational guides from third-party sources dominate AI citations, with brands mentioned as examples rather than as primary sources.]

The practical implication: AI SEO strategy stops being about answering every question and starts being about answering specific questions better than anyone else. That’s a meaningful shift in how content is briefed, produced, and measured. Good AI keyword research makes that brief concrete, identifying exactly which topics and contexts to prioritize.

Three Actions That Close the Gap

The citation audit produces a specific output: for each topic cluster and each platform, it identifies which type of action is most likely to close the visibility gap. Those actions fall into three categories, each with different resource requirements and timelines.

Digital PR Owned content Social / UGC
Earn third-party mentions Partner with publishers AI draws on. Contribute expert commentary. Be included in sector guides. Build authority content Comprehensive guides, comparison pages, original data. Topics the audit identifies as underserved. Community presence Be credible where buyers research before reaching your site. Longest runway, growing signal weight.
Fastest impact Citations driven by external mentions, not owned pages Medium-term Depends on topic gap size and content quality Longest runway Matters increasingly as AI incorporates social signals

Digital PR and third-party mentions are the highest-leverage activity for most brands, because they address the most common finding: that the majority of AI citations are coming from independent sources, not owned pages. The goal is to be embedded in the content ecosystem for your topic. That means partnering with the publications, advisory firms, and consultancies that are producing the content AI draws on. Contributing expert commentary, providing authoritative reference material that others can link to, and collaborating on guides where your brand appears as a contextual example alongside competitors. 

Owned content investment is the right response when the citation audit shows that your owned pages are genuinely absent from the topic, not just outperformed. The priority isn’t more content; it’s better content in the right areas. The audit identifies exactly which topics are underserved. The content itself needs to be the type that AI systems and third-party sites can cite: comprehensive guides that cover a topic with real depth, comparison pages that place your offer in context, step-by-step process guides built around specific use cases, and, where possible, original data or analysis that doesn’t exist elsewhere. Depth and specificity earn citations. Breadth and volume don’t.

Social and community presence is the response when visibility gaps are driven by UGC signals, typically in topics where buyers seek peer experience and independent comparison rather than brand-produced content. Community management in the right channels, credible participation in conversations on Reddit, Quora, and industry forums, and authentic engagement rather than promotional presence. This is the longest runway of the three, but it’s growing in importance as AI systems increasingly incorporate social signals into what they surface.

The Bigger Picture: Presence Over Position

Traditional search was about position. Rank highly, earn traffic, convert at the margin. Visibility was a number: position one, page one, top ten. You knew where you stood, and you optimized to move up.

AI-driven search works differently. A brand can shape what users learn about a category, influence the answer to a high-intent question, and be present at the moment a decision is forming, all without appearing as a link. Visibility is no longer a rank. It’s a probability: how likely are you to be present when it actually matters?

The brands that understand this earliest are building an advantage that compounds. Not because they’ve found a new SEO trick, but because they’ve shifted their content investment toward genuine authority in specific contexts, and that authority is what AI systems consistently draw on.

That’s the conclusion the citation audit points to, and it’s what makes AI visibility tools genuinely useful when they’re used right. They serve as a diagnostic that tells you where authority is missing and what to build next.

Success in this environment is defined by presence, not position. The content strategy implications follow directly from that.

FAQs

How do you audit AI search optimization response analysis?

Start by running structured prompts across the major AI platforms, covering the topics most relevant to your buyers’ decision-making process. Analyze which pages are being cited in responses to those prompts, and categorize them by source type: third-party, owned, or social. The distribution tells you where the gap is coming from and what type of action closes it. Secondary metrics, including run length, entropy, and Gini coefficient, reveal how stable your visibility is and how competitive each topic is.

How do you use AI for a content audit?

An AI citation audit is a specific type of content audit that goes beyond traditional performance metrics. Rather than measuring traffic or rankings for your owned pages, it measures how often your brand and content appear in AI-generated responses to relevant prompts. The output identifies which topics are underserved, which content types earn citations, and whether the gap requires digital PR, new owned content, or community presence. It connects content decisions directly to AI visibility outcomes.

 How do you audit for AI search visibility?

Build a structured set of prompts using the SPIV framework, grounded in your actual buyer personas and intent stages rather than generic category terms.

Pair that with AI keyword research to identify the topic gaps the audit surfaces, and you have a complete workflow from measurement to action.

Run those prompts across ChatGPT, Google Gemini, Perplexity, and Google AI Overviews on a recurring basis. Track both primary metrics from the platform and secondary metrics calculated on top of the export data. The citation analysis, which identifies what sources AI is drawing on and where your brand appears in that ecosystem, is the layer that tells you what to do next.

Conclusion

This series started with a measurement problem.

Most teams tracking AI visibility are using deterministic tools to measure a probabilistic system, running generic prompts that describe buyers who rarely exist in practice. The data looks clean. The picture it paints isn’t representative.

The response to that problem was a methodology: structured prompt construction grounded in real buyer personas and intent stages, a two-layer metric system that separates surface-level visibility from genuine diagnostic insight, and a modular audit format that makes the output actionable rather than overwhelming.

What the citation audit adds to that is the strategic implication. AI visibility is built primarily through third-party mentions, not owned pages. Coverage-first content is the most exposed to displacement. Genuine authority in specific, high-intent contexts is what earns consistent citations. The content investment that follows from that is about producing the right things, in the right depth, for the contexts where decisions actually happen.

The brands that make that shift now will hold ground as search continues to change. The ones that don’t will keep producing content that looks healthy in their dashboards while becoming invisible in the moments that matter most.

Read more at Read More

How We Rebuilt AI Visibility Measurement From The Ground Up

/0 Comments/in /by

Key Takeaways

  • The core problem with most AI visibility prompts isn’t that they’re wrong; it’s that they’re missing the context real users bring. Generic inputs produce generic, unactionable data.
  • The SPIV framework (Segment, Persona, Intent, Variable) structures prompts around four variables drawn from real user data, turning stateless AI visibility tracking inputs into high-fidelity user proxies.
  • Once prompts are grounded in real context, the variation you observe in model responses becomes informative rather than noise. Visibility can then be expressed as a probability distribution.
  • Measurement operates on two layers: primary metrics from the tracking platform, and a secondary layer of calculated metrics (run length, Shannon entropy, Gini coefficient, and KL divergence) that reveal the stability and competitive dynamics behind the surface numbers.
  • This approach naturally connects measurement to business priorities. It becomes much harder to justify tracking low-intent queries with no connection to how your product is actually bought.

The first post in this series made the case that most AI visibility tracking is built on the wrong foundation: generic prompts measuring hypothetical users, deterministic tools applied to a probabilistic system. If that diagnosis is right, the obvious next question is: what does a better approach actually look like?

That’s what this post covers. What we built at NP Digital to address both the measurement problem and a second issue that compounded it: early AI visibility audits were trying to do too much at once, producing outputs so dense that clients couldn’t identify a single clear action to take. The rebuild addressed both problems together.

The result is a methodology built around structured prompt construction, two layers of metrics, and outputs that point to specific, defensible actions. Here’s how it works.

Why the Old Audit Approach Wasn’t Working

Before explaining what we built, it helps to explain what we were moving away from, and why.

Early AI visibility audits, including our own initial attempts, were structured like SEO audits. A single document tried to cover everything at once: a content audit, a competitor audit, a structured data review, citation analysis, and strategic recommendations, all bundled into one output. The logic made sense at the time. SEO audits had always worked this way. Why would a GEO audit be different?

The answer, in practice, was that clients couldn’t use them. Data points conflicted. The strategic direction wasn’t clear. The same document had to be re-presented multiple times before anyone could agree on what to do first. We were producing thorough work that left clients more confused than when they started.

Two problems were running in parallel. The first was the measurement problem I covered previously: generic prompts producing data that looked meaningful but wasn’t representative of real buyer behavior. The second was a presentation problem: even if the data had been better, the format buried the signal in too much noise.

A comparison of different approaches to building topic clusters.

The rebuild addressed both. On the measurement side, we moved to structured prompt construction through the SPIV framework. On the output side, we separated the analysis into discrete, digestible pieces: each focused on a specific topic cluster, each pointing to a defined type of action. Clients stopped needing multiple sessions to understand what they were looking at.

Introducing the SPIV Framework

The starting point is familiar data. The same sources that feed traditional keyword research, including People Also Ask results, Google Search Console data, community platforms like Reddit and Quora, and first-party data like customer service transcripts where available, provides the raw material. The difference is what happens next.

Instead of using those inputs as-is, SPIV treats them as raw material and injects four structured variables into each prompt. The practical effect: it turns stateless AI keyword research inputs into pseudo-stateful responses by giving the model the persona context it would otherwise be missing.

The S.P.I.V.framework explained.

Each variable does a specific job:

  1. Segment: The market category or business context. Grounds the prompt in a defined situation: ‘SME owner in the UAE’ rather than ‘business owner.’ This is the broadest layer of context.
  2. Persona: The specific user type, including relevant traits: risk tolerance, level of prior knowledge, geographic or professional context. This is where abstract ‘users’ become real people with real constraints.
  3. Intent: What the user is actually trying to accomplish, not the topic they’re searching but the outcome they need. ‘Understand my compliance obligations’ is different from ‘find the cheapest option.’ Separating these surfaces meaningful differences in how models respond.
  4. Variable: A single modifier that can be shifted to test sensitivity: ‘fastest’ vs. ‘cheapest’ vs. ‘most reliable.’ Isolating one variable at a time makes the data interpretable. Change everything and you can’t explain what moved.

The table below shows what this transformation looks like in practice, using anonymized examples from real audit work:

A prompt optimization metrics for AI visibility audits.

The difference between the raw input and the SPIV-optimized prompt isn’t cosmetic. The raw prompt describes no one in particular. The optimized prompt describes a specific person in a specific situation trying to accomplish a specific outcome. That specificity is what makes the model’s response meaningful as a measurement input.

A well-constructed set of SPIV prompts doesn’t need to be large. Representativeness matters more than volume. A focused set of 15 to 30 prompts mapped to your key buyer personas and intent stages gives more actionable signal than hundreds of generic variations.

The Two Layers of Measurement: Primary and Secondary Metrics

Once prompts are properly constructed, the analysis operates on two distinct layers. Understanding the difference between them is what makes the output useful rather than just interesting.

Primary metrics come from the tracking platforms directly, including Writesonic and Profound. These include visibility percentage, share of voice, and mention frequency. They’re the standard outputs most teams are already familiar with and they provide the baseline picture: how often does your brand appear, and how does that compare to competitors?
 
The four secondary metrics, and what each one tells you:

  1. Run length: The number of consecutive days a brand maintains visibility for a given topic. Short run lengths signal volatile, unreliable presence. Long run lengths indicate that the model has formed a stable association between the brand and that topic, what we’d call persistent authority rather than a transient mention.
A guide to interpret run length in an AI visibility edit.
  1. Shannon entropy: A measure of how evenly visibility is distributed across the brands appearing for a given topic. High entropy means no brand dominates, meaning the model is pulling from a wide, fragmented field. Low entropy means the results are concentrated, and that a small number of brands are taking most of the mentions. Low entropy topics are harder to break into; high entropy topics are more contestable.
  2. Gini coefficient: Where Shannon entropy tells you how distributed results are, the Gini coefficient tells you the degree of concentration. A high Gini score means visibility is dominated by one or two brands. A low score means the field is relatively open. Together with entropy, this gives a picture of whether a topic is winner-takes-most or genuinely shared.
A chart to interpret the Gini coefficient in an AI visibiity edit.
  1. KL divergence: In a traditional statistical context, this metric measures how a distribution changes over time. We’ve adapted it here to serve a different purpose: measuring how far an individual platform’s results drift from the group average across all tracked platforms. A low score for a given platform means its brand rankings for that topic are broadly in line with the consensus across ChatGPT, Gemini, and Perplexity. A high score means that platform is picking a significantly different set of brands. That’s a meaningful finding. It tells you whether your visibility is genuinely broad or whether it’s concentrated in one model’s view of the world.
A guide on interpreting KL divergence for AI visibility edits.

None of these metrics is useful in isolation. Run length tells you how stable your visibility is; entropy and Gini tell you how competitive the topic is; KL divergence tells you whether that visibility holds across platforms or is fragile in a way your headline numbers don’t reveal. Read together, they give a diagnostic picture that primary metrics alone can’t produce.

What the Data Tells You

With SPIV-structured prompts and both metric layers in place, visibility stops being a single number and becomes a probability distribution. The question changes from ‘where do we rank?’ to ‘how reliably do we appear when the conditions that actually matter are present?’

In practice, this approach surfaces findings across three dimensions that generic tracking misses entirely.

The visibility distribution itself. Some brands are category staples: they appear consistently across multiple runs of the same prompt, across slight variations in phrasing, across different platforms. Others are volatile outliers: they surface occasionally but can’t be relied on. Generic tracking averages this out and produces a headline figure that obscures the difference. The secondary metrics separate the two clearly.

A graphic explaining how visibility should be defined when it comes to AI/LLMs.

The platform dimension. Visibility that holds on Google Gemini but not on ChatGPT is a meaningful finding, not just a data point to average away. Different models draw on different training data, weigh different source types, and respond differently to the same underlying intent. KL divergence makes this visible. A brand that appears strong in aggregate but has a high divergence score on one platform has a concentration risk that matters strategically, especially if that platform is where your buyers actually research.

The topic dimension. This is often the most strategically important finding in the whole audit. Brands regularly show strong visibility in broad, low-intent queries (the general category terms that show up well in standard tracking), but near-zero presence in the specific, high-intent topics their buyers are researching at the point of decision.

In one audit, a brand showed visibility above 65 percent for general licensing topics across platforms. For compliance and banking topics (the two areas most directly connected to their buyers’ decision-making process), visibility was zero across ChatGPT, Google AI Overviews, and Perplexity. The standard tracking looked healthy. The actual picture was that the brand was invisible at the moments that mattered most.

Generic prompts miss this because they aren’t asking the right questions. SPIV-structured prompts surface it because they’re built around the contexts where decisions actually happen.

This is also where the measurement connects directly to AI SEO strategy. Once you know which topics show gaps, which platforms are most divergent, and which competitors are holding the positions you’re not, you have a defensible brief for content and PR investment. The audit doesn’t just tell you where you are. It tells you where to go.

FAQs

How do you track AI visibility?

Tracking AI visibility starts with a defined prompt set run across the major platforms: ChatGPT, Google Gemini, Perplexity, and Google AI Overviews. Tools like Writesonic and Profound automate this process and export visibility data by brand and topic. The critical step most teams skip is structuring those prompts around real buyer personas and intent contexts rather than generic category terms. Generic prompts produce directional data; structured prompts produce data you can act on.

How do you monitor brand visibility in AI?

Brand visibility in AI is monitored by running structured prompts across platforms on a recurring basis and tracking both primary metrics (visibility percentage, share of voice) and secondary metrics (run length, entropy, Gini coefficient, KL divergence). The primary metrics tell you what the numbers are. The secondary metrics tell you whether those numbers are stable, how competitive the topic is, and whether your visibility is genuinely broad or concentrated on a single platform. Monitoring both layers gives you a picture you can act on.

How do I check AI visibility of my brand?

Start by identifying the topics most relevant to your buyers’ decision-making process, not just the broad category terms, but the specific questions they ask when they’re close to a purchase. Build prompts around those topics using the SPIV framework, run them across ChatGPT, Gemini, Perplexity, and Google AI Overviews, and track how consistently your brand appears. The gap between your visibility in general topics and your visibility in high-intent, decision-stage topics is usually the most important finding.

Conclusion

The shift this methodology makes is simple to state but significant in practice: you’re no longer tracking where you rank. You’re tracking how reliably you appear when it actually matters: for the right persona, at the right intent stage, on the platforms your buyers actually use.

SPIV is how you build the inputs that make that measurement possible. The secondary metrics are how you make sense of what the data is telling you. Together, they turn AI visibility from a headline number into a diagnostic that points somewhere useful.

Knowing where you’re visible and where you’re not is only half the equation. In the final post in this series, I’ll cover what this framework reveals about content strategy, and why the old volume-first approach doesn’t hold up in an answer-driven search environment.

Read more at Read More

AI Brand Visibility: You’re Tracking It Wrong

/0 Comments/in /by

Key Takeaways

  • Most AI brand visibility tracking today replicates keyword tracking logic, using prompts instead of search terms. The underlying assumption is the same, and that’s the problem.
  • Traditional search engines are deterministic: the same query tends to return similar results. LLMs are probabilistic: the same prompt can produce a wide range of valid answers.
  • Measuring a probabilistic system with deterministic tools produces data that looks clean but doesn’t reflect how the system actually behaves.
  • The prompts most brands are tracking (‘Best CRM in 2026,’ ‘Top accounting software’) describe a user who doesn’t exist, someone with no context, no history, and no specific intent. This is a known gap in current AI SEO measurement approaches.
  • Fixing this requires a different measurement philosophy, not just better prompts.

Have you started tracking your brand in ChatGPT, Perplexity, or Google AI Overviews? Good. You’re thinking about the right problem.

Here’s the harder question: what are you actually measuring?

Most teams doing AI brand visibility tracking today have taken a familiar mental model and applied it to an unfamiliar system. Prompts have become the new keywords. Visibility scores have become the new rankings. Tracking platforms have emerged to show how often your brand appears in AI responses over time. On the surface, it looks like a natural evolution of the work you’ve already been doing.

It isn’t.

The tools built for traditional search were designed for a deterministic system, one where the same query reliably returns the same results. Large language models (LLMs) don’t work that way. They’re probabilistic: the same prompt can produce a range of valid answers, shaped by phrasing, context, model version, and more. Applying rank-tracking logic to a system that doesn’t produce ranks is the core mismatch, and it’s quietly corrupting the data most teams are reporting on.

This post breaks down exactly what’s going wrong and what a better approach looks like. It’s the first in a three-part series on AI visibility measurement. Part two introduces a structured framework for building prompts that actually reflect how your buyers use AI. Part three covers what the resulting data reveals about your content strategy.

The Tool The Industry Reached For (and Why It Doesn’t Fit)

The industry’s current approach to AI visibility measurement wasn’t irrational. It was fast. When a new channel emerges, teams reach for the tools and frameworks they already understand, and in digital marketing, that means rankings, share of voice, and tracked keywords. The logic was simple: prompts are the new search queries, so treat them the same way.

The problem is that search engines and LLMs are fundamentally different types of systems.

Traditional search is deterministic. Submit the same query to Google twice and you’ll get a broadly similar set of results. Position may shift slightly, but the system is stable enough that rank tracking works. That predictability is the entire foundation of AI keyword research and traditional SEO measurement.

LLMs are probabilistic. Run the same prompt multiple times and you’ll get a distribution of responses, not a fixed answer. The model generates each response based on statistical associations, not a retrievable index. There is no ‘rank one’ to hold.

The table below illustrates the mismatch. Applying rank-tracking logic to a probabilistic system doesn’t give you a less accurate version of the right answer. It gives you a fundamentally different kind of measurement entirely.

  Traditional Search LLM Ecosystem
System Type Deterministic Probabilistic
Behavior Predictable / Stable Variable / Generative
Core Metric Rank (Position) Presence (Likelihood)
Same query = same result? Broadly yes Not necessarily

This isn’t a minor calibration issue. It’s structural. If you’re reporting on AI visibility using methods designed for predictable, stable systems, you’re building strategy on a foundation that doesn’t reflect how LLMs actually work.

The User Who Doesn’t Exist

The second flaw in current AI visibility tracking is less obvious but equally important.

Most prompt tracking today relies on generic, decontextualized inputs:

  1. ‘Best CRM in 2026’
  2. ‘Top accounting software’
  3. ‘Best project management tool for small teams’

These prompts are clean, scalable, and easy to standardize. They look exactly like the keywords we’ve always tracked.

They also don’t resemble how real people use AI tools.

Real users carry context. They have prior conversations, professional constraints, specific goals, and levels of knowledge that shape what they’re actually asking. A prompt like ‘Best CRM in 2026’ represents an abstract, anonymous user with no history, no constraints, and no intent beyond the words in the query.

A graphic breaking down the differences between abstract users and how actual users use LLMs.

So when you measure AI visibility using these prompts, you’re measuring how the model responds to a hypothetical person who rarely shows up in real decision-making moments. That’s directionally useful at best.

Real audit work bears this out. In one analysis, a brand showed strong visibility for broad category queries, the kind that show up well in standard tracking. But when prompts were shaped around the specific contexts their buyers actually operate in, visibility dropped to zero in the topics most directly connected to purchase decisions. The tracking looked healthy. The actual picture wasn’t.

Generic prompts measure AI visibility for a user who rarely exists. If you want to know how your brand appears to real buyers, you need inputs that reflect real buyer contexts.

The Scaling Trap

The instinctive response to ‘generic prompts aren’t representative’ is volume. If one prompt isn’t enough, run a thousand variations. Add synonyms, modifiers, intent signals, geographic qualifiers. Cover the space more thoroughly.

This logic leads directly into what we call the scaling trap.

Every topic branches into multiple phrasings, intents, personas, and contextual modifiers. The number of prompts required to meaningfully approximate reality grows exponentially. A topic with five main phrasings, three intent signals, and four persona types generates 60 prompt combinations before you’ve added geographic variation or industry context. Scale that across a full content strategy and you’re looking at tens of thousands of prompts, run repeatedly, across multiple models, on a recurring basis.

A graphic explaining the volume fallacy and how prompts properly reflect reality.

Two problems follow. The first is practical: the cost of running this at scale is significant, and it compounds across every client account and every reporting cycle. The second is more fundamental: even after all of that, there’s no guarantee the resulting dataset is meaningfully more representative of actual user behavior. You’ve scaled the volume without fixing the flaw in the input logic.

More prompts don’t fix a representativeness problem. They just make the flawed measurement more expensive.

What Good Measurement Actually Requires

If the problem is that prompts lack context, and brute-force volume doesn’t solve that, the answer is to improve the quality of the input rather than the quantity.

Good measurement of a probabilistic system requires asking a different question entirely. The old question was: ‘Where do we rank?’ The right question is: ‘How reliably does our brand appear when the conditions that actually matter are present?’

That shift has real implications. A brand that appears 85 percent of the time when the right persona and intent conditions are met has a genuinely strong position, even if its average visibility across generic prompts looks modest. A brand that appears 50 percent of the time on generic queries but near zero percent in high-intent, decision-stage contexts has a problem that average tracking completely obscures.

Visibility, measured correctly, is a probability distribution across specific user contexts, not a single score. Getting to that measurement requires inputs that reflect those contexts: structured prompts built around real user personas, specific intent stages, and the actual questions buyers ask when they’re close to a decision.

That’s the foundation of a better approach to AI visibility measurement. The next post in this series walks through exactly how to build it.

Image related to AI Brand Visibility: You’re Tracking It Wrong

In the next post, I’ll walk through the framework we use at NP Digital to build prompts that reflect how real buyers actually engage with AI and what the data looks like when you do it right.

Why This Matters Now

AI-driven search has moved from a future consideration to a present reality, faster than most marketing teams anticipated.

ChatGPT now has over 700 million users, with exponential growth going on. That’s not a niche research tool. That’s a primary discovery channel for a significant and growing share of your buyers.

Image related to AI Brand Visibility: You’re Tracking It Wrong

Google AI Overviews now appear on roughly 48 percent of tracked queries, up 58 percent year over year according to BrightEdge data. In B2B technology, that figure reaches 82 percent of queries. If your buyers research software, services, or professional categories, AI is already shaping what they find before they ever reach your site.

The competitive dynamics are shifting accordingly. Brands that appear consistently in AI responses for the right queries, at the right intent stages, are building an advantage that compounds over time. Brands that don’t appear, or that appear for the wrong queries, are losing ground in the consideration phase before a sales conversation ever starts.

Every week you’re tracking AI visibility with flawed inputs is a week you’re making content and strategy decisions based on data that doesn’t reflect how your buyers actually use AI. The window to get ahead of this is open now.

FAQs

Why should I track AI brand visibility?

Your buyers are already using AI tools to research options, compare solutions, and form opinions about your category. Tracking AI brand visibility tells you whether your brand is present in those moments or invisible. Unlike traditional search, where a low ranking is visible and actionable, AI invisibility is silent, so you won’t know it’s happening unless you measure it.

What Is AI visibility?

AI visibility refers to how often and how favorably your brand appears in responses generated by AI tools like ChatGPT, Perplexity, Google Gemini, and Google AI Overviews. Strong AI visibility means your brand is being surfaced when users ask questions relevant to your product or service.

What are the top AI visibility solutions?

The most widely used platforms for tracking visibility include Writesonic and Profound, alongside a growing number of specialist tools. Each uses a defined prompt set to measure how often your brand appears across major AI platforms. The quality of your prompt set determines the quality of what you can learn — which is exactly the problem I want to address with this series.

Conclusion

Marketers aren’t doing something foolish by tracking AI visibility. They’re doing something natural: applying the tools and mental models they already know to a new channel. The problem is that those tools were built for a deterministic world, and LLMs don’t operate that way.

The mismatch matters. It means the data most teams are reporting on is structurally limited, not wrong exactly, but not representative of what’s actually happening when your buyers use AI to research your category.

The fix starts with a different question. Stop asking where you rank. Start asking how reliably you appear when it actually matters.

In the next post in this series, I walk through a framework built specifically for that question. This is a structured approach to prompt construction that reflects real buyer contexts and makes probabilistic measurement genuinely useful.

Read more at Read More

Bing Webmaster Tools updates AI reporting with Intents, Topics, Citation Share and Compare

/0 Comments/in , /by

Microsoft is now officially releasing a preview of the new AI performance report within Bing Webmaster Tools that now includes Intents, Topics, Citation Share, and Compare. We saw Microsoft demo these features in late April but now it is actually starting to roll out to users.

As a reminder, Bing officially rolled out its AI performance report in February. Google didn’t roll out its AI reporting in Search Console until June, and it seemed forced.

What is new. “These new capabilities build on that foundation by helping publishers better understand why their content is being surfaced, which broader subject areas they are gaining visibility in, how their presence evolves relative to other cited sources, and how citation patterns change over time,” wrote Krishna Madhavan from Microsoft.

Intent: The new Intents feature in Bing Webmaster Tools now classifies the grounding queries in the AI Performance Report in broader categories, such as Informational, Commercial, Navigational, Learn and Solve, Research, Creation, Local, and more. This in a sense helps you understand the intent behind the prompt or query. “This helps publishers move beyond simply seeing which queries triggered citations and begin understanding the broader query context our systems associate with those citation appearances,” Krishna Madhavan wrote.

The example provided was that an e-commerce publisher may discover strong visibility in comparison-oriented or shopping-focused AI experiences, while an educational publisher may find that their content is frequently surfaced in research or learning-oriented interactions. These insights can help publishers better align content structure and depth with the types of experiences where AI systems are surfacing their content.

Topics: The Topics in the AI performance reports group related grounding queries into broader thematic clusters. AI systems reason across concepts and themes rather than isolated keywords, Microsoft explained. So by having topics, it will help publishers understand visibility in the same thematic structure that modern AI systems use to organize information.

So for example, queries such as “solar panels,” “solar energy efficiency,” and “residential solar installation,” for example, may all map into a broader topic cluster like Solar Energy.  “This creates a more natural way to analyze AI visibility. Content teams and publishers often think in terms of themes, editorial areas, and audience interests rather than isolated keywords. Topics help bridge that gap by turning grounding query data into a more thematic view of AI engagement,” Microsoft wrote.

One note, “during the preview phase, some labels may still be broad – especially for highly specialized or niche domains – but the system is already beginning to reveal meaningful thematic patterns,” Microsoft wrote.

Citations. Microsoft also added citation share, which shows how much of the citation space your site receives for a specific grounding query. Citation share is calculated as the percentage of citations attributed to your site out of all citations shown across all sites for that same grounding query. “This helps publishers understand not just whether they were cited, but how much visibility they received within the full set of cited sources for that query,” Microsoft explained.

Microsoft added these points:

  • “This can provide a more directional view into how visibility is evolving over time. Publishers may begin to identify areas where their content has strong and growing representation in AI-generated experiences, as well as areas where visibility may be more fragmented across many sources.”
  • “Importantly, Citation Share is designed as an observational metric – not a ranking system or a competitive scoreboard. It does not expose competitor domains, represent traffic share, or assign quality scores to content.”
  • “AI citation ecosystems are inherently dynamic. Citation patterns can shift due to changes in user behavior, evolving models, freshness signals, partner refresh cycles, and broader changes across the web itself.”

Compare. With all of that, you can also compare the changes over time. The compare feature allows you to overlay a previous time period directly onto the current reporting view. 

“Compare is designed to help publishers observe changes over time. Citation activity can be influenced by many factors including evolving AI models, competing content, freshness signals, and shifts in user demand,” Microsoft wrote.

Here is what it looks like:

Why we care. While we still do not have click and click-through rate data, Microsoft keeps adding more and more to its AI performance reports.

I am hopful that one day we will get click data, but I am still not expecting to see that from Google or Microsoft any time soon.

Read more at Read More

Google penalties: Why prevention is cheaper than recovery

/0 Comments/in , /by
Google penalties- Why prevention is cheaper than recovery

Google penalties, also known as manual spam actions, are among the few events in search that can disrupt an otherwise healthy online business overnight. 

For companies heavily dependent on organic traffic, the consequences often extend far beyond lost rankings. Revenue drops, customer acquisition costs rise, expansion plans stall, and the effects can linger long after the original policy violations have been remedied.

With a steady 90% market share, Google remains the primary traffic source for many publishers, ecommerce platforms, retailers, travel brands, affiliates, and lead generation businesses. 

Direct traffic rarely compensates for a major visibility loss, and Bing seldom offsets the difference. As a result, a manual spam action carries serious operational implications, not merely SEO risks.

Manual actions aren’t algorithm updates

One point still misunderstood throughout the industry deserves clarification. Manual spam actions differ from algorithmic updates. They aren’t fluctuations caused by changes in relevance calculations or ranking system adjustments. 

Google’s manual penalties involve direct enforcement after suspected violations against Google Search Essentials, formerly Google Webmaster Guidelines, have been identified and confirmed. The distinction matters because the response required is completely different.

A website affected by changing ranking systems requires analysis, adaptation, and recrawling. A website affected by a manual spam action requires remediation and applying for reconsideration. Those are separate situations entirely.

Google doesn’t issue manual spam actions casually. The process involves internal senior employee review cycles. Suspected violations must be investigated and confirmed first.

Google states clearly that manual actions are the consequence of proven policy transgressions. Despite frequent cries of foul, false positives are exceptionally rare. Once a manual action appears in Google Search Console, the enforcement is already in the production pipeline.

The operational problem is that many businesses fail to recognize how much unresolved policy exposure their web platforms have accumulated over time.

Be the brand customers find first.

Track, grow, and measure your visibility across Google, AI search, social, local, and every channel that influences buying decisions.

Start your free trial

How penalties develop

The initial steps that ultimately lead to a manual penalty and a website’s drop in search visibility often begin inconspicuously, gradually eroding policy compliance. 

  • An ecommerce business launches an aggressive link acquisition campaign during an early growth phase. Over the years, PageRank-passing spam links accumulate unchecked until eventually nobody remembers where thousands of exact-match backlinks originate. 
  • A publisher enters into commercial partnerships involving sponsored content or affiliate sections, which gradually become structurally integrated into the editorial architecture of the website. 
  • A SaaS company creates large numbers of low-quality location pages while expanding into new markets. 
  • A lead generation business scales supplemental SEO content through low-cost LLM production systems with limited editorial oversight because that appears to be what most competitors are doing.

The underlying patterns are remarkably similar across industries. In many cases, organic search visibility initially improves and may even generate measurable revenue gains attributable to the SEO initiative. 

The short-term results reinforce the perception that the approach is working. However, as time passes, nobody revisits whether those earlier decisions remain aligned with evolving search quality standards and webmaster policies.

Why historical violations still matter

One reason manual spam actions create so much disruption is that policy violations often persist quietly for years before review. Many organizations incorrectly assume that questionable SEO tactics of the past lose their relevance over time. 

Yet Google Search systems don’t forget historical footprints. Archived URLs remain crawlable. Legacy sections continue contributing content quality signals long after internal ownership was abandoned. 

Most persistently, backlink patterns remain visible for decades. Large numbers of websites remain affected by backlinks generated through manipulative campaigns dating back many years. 

Paid placements, article syndication networks, private blog networks, commercial keyword-heavy guest posting campaigns, expired domain backlinks, directory spam, and widget distribution schemes that once formed part of mainstream SEO activity are today’s liabilities. 

Some of these practices continue to operate more or less openly for years, while enforcement may appear erratic or inconsistent. When left unaddressed, they represent an incalculable risk to the website publisher.

Get the newsletter search marketers rely on.

See terms.

Time doesn’t eliminate the risk

This becomes particularly important during acquisitions. Businesses purchasing established domains frequently inherit unresolved compliance exposure alongside rankings and traffic. Google evaluates the website’s condition, not which employee, agency, or previous owner introduced the violations.

Traffic growth alone doesn’t confirm compliance health. A domain generating millions of clicks may still carry unresolved risks tied to old link schemes, expired sponsorship arrangements, deceptive user-agent cloaking, manipulative redirects, or scaled low-quality content sections. Those issues often go unnoticed until they’re brought to the surface by a Google manual spam action notification.

A common sign of a manual spam action: rapid loss of visibility
A common sign of an algorithmic adjustment: Gradual loss of visibility

Reputation abuse and publisher liability

The mechanics behind reputation abuse are straightforward. A trusted brand with an established web platform allows third parties to publish unrelated, often unsupervised content under the same domain name. In many cases, publishers integrated discount coupon sections, casino reviews, affiliate content, or commercially motivated informational pages directly into existing editorial systems. 

The problem frequently worsened significantly because the content wasn’t properly segmented. The consequence is that the distinction between trusted editorial work and commercially motivated material became blurred.

Once confronted with a site-wide penalty, affected publishers experience broad visibility declines across the entire platform, not merely within the originally offending sections of the website. The damage to a brand that lends its reputation to a disreputable third party is often substantial. 

Recovery efforts frequently prove time-consuming and costly. Removing isolated pages rarely resolves the problem. Many organizations require broader structural changes, including archive cleanup, internal link reviews, crawl management adjustments, sponsorship governance reforms, the removal of spammy redirects, stronger editorial oversight, and stricter technical segmentation.

In short, recovering from such a penalty takes time, costs significant amounts of money, and is often a painful process.

A common sign of a manual spam action: Rapid loss of visibility
A common sign of a manual spam action: Rapid loss of visibility

The risks of scaled content

Google increasingly scrutinizes large-scale publishing systems that produce repetitive, low-value content without a unique selling proposition. 

The issue isn’t maintaining many websites simultaneously. Large website portfolios have thrived in Google Search for years and continue to do so. The underlying problem involves quality control, editorial oversight, originality, and informational value.

  • Affiliate networks produce near-identical product comparison pages across thousands of long-tail keywords. 
  • Local SEO operations deploy templated service pages across hundreds of regions with minimal differentiation. 
  • AI-assisted workflows publish large numbers of informational pages without factual oversight or genuine expertise to support them. 
  • Travel websites generate mass-produced destination pages through repetitive, generic content systems.

Most organizations don’t cross into problematic territory intentionally. The transition usually occurs gradually, often unbeknownst to the decision-makers who rely on outdated or misleading recommendations. 

The resulting manual spam action in Google Search Console, followed by a sharp decline in rankings, frequently occurs after a prolonged period of spam signal accumulation rather than during the apparent growth phase.

Incomplete remediation prolongs penalties

Many site owners approach reconsideration requests as if they were negotiating with Google. That puts them at a significant disadvantage from the outset. 

The reconsideration process exists for one purpose only: to demonstrate that the website has been restored to full compliance with Google’s guidelines. It’s important to note that Google expects complete compliance before lifting a manual spam action.

This means the requirement extends beyond the specific violation highlighted in Google Search Console. A site owner who addresses only one known spam issue while leaving unrelated policy violations unresolved elsewhere on the website will typically face rejection.

A common testing approach, such as a publisher removing some problematic sponsored content while retaining similar affiliate arrangements elsewhere, will fail. Likewise, a business that disavows recent manipulative backlinks while ignoring historical paid link schemes is unlikely to convince Google of its genuine commitment to complying with Google’s policies going forward.

Similarly, a website network that cleans up one property while continuing identical publishing practices across related domains signals incomplete remediation rather than meaningful operational reform. As a result, it stands little chance of regaining Google’s trust.

Why repeated rejections make recovery harder

Effective website recovery requires a comprehensive review rather than selective cleanup. Technical infrastructure, content quality, sponsorship structures, redirect behavior, link acquisition history, indexing patterns, archive sections, and ownership transparency all require examination during serious compliance recovery efforts.

The Google Search team expects compelling documentation detailing what has changed and how future violations will be prevented. Temporary cosmetic adjustments rarely persuade reviewers to lift a manual spam action.

Making matters worse, each rejection typically requires an even more comprehensive review and cleanup effort. At the same time, every reconsideration request that Google deems disingenuous further erodes Google’s trust in the publisher.

The cost of uncertainty

There’s no guaranteed turnaround time for reconsideration processing. Some reviews are completed within days. Others take weeks or months.

At the same time, large websites with extensive SEO legacies accumulated over many years often require longer assessment periods due to the substantial volumes of data that must be crawled and analyzed before changes can be evaluated.

For businesses that rely primarily on Google traffic, this uncertainty creates a potentially existential threat.

  • An ecommerce business approaching a peak seasonal period with an unresolved manual spam action can face cash flow problems quickly.
  • Publishers dependent on advertising revenue experience ranking losses that translate directly into declining commercial performance.
  • Lead generation businesses often encounter immediate pipeline contraction once visibility declines significantly.

The operational risk becomes even greater when companies fail to build a strong brand capable of partially offsetting organic traffic declines through direct navigation or alternative revenue-generating channels. In this context, paid traffic is a poor substitute due to its associated costs.

In short, some online businesses can’t afford to be penalized in the first place.

Penalties can cripple operations

The issue extends beyond SEO performance. Search visibility directly affects commercial expansion, investor confidence, company valuation, partnership negotiations, and revenue stability.

Penalty expiration represents another commonly misunderstood aspect. Google manual spam actions may expire after prolonged periods, often years. However, this is rarely a viable strategy for an affected business. 

Waiting passively through an extended period of declining visibility seldom aligns with commercial realities. More importantly, expiration alone doesn’t guarantee recovery or renewed growth, as the penalty could be reapplied not too long after it expired.

Google’s search systems continue evaluating overall site quality independently of manual enforcement status. A website carrying unresolved spam signals across its content, technical infrastructure, or off-page profile may continue to struggle long after the manual action itself has been lifted.

Compliance requires ongoing oversight

Compliance reviews can’t be considered optional or a luxury. Organizations heavily dependent on organic Google visibility require ongoing operational review cycles focused specifically on comprehensive policy compliance.

These reviews shouldn’t be conducted internally. Even the most talented in-house SEO teams are often hard-pressed to diligently identify shortcomings that may reflect on their own work or that of their colleagues. Policy compliance requires external expertise, sufficient authority, and a proven track record.

Purely technical SEO audits, while indispensable, are insufficient if commercial partnerships bypass oversight. Editorial standards alone won’t suffice if historical link manipulation remains unresolved. Planned growth initiatives require evaluation against established compliance frameworks before deployment, not after traffic has become dependent on questionable practices.

Mature organizations increasingly integrate compliance reviews into their operational governance. Sponsorship structures undergo search compliance review before launch. Scaled publishing systems are assessed for quality before expansion. Historical content is evaluated on a recurring basis. Acquisition due diligence includes policy exposure analysis alongside financial review.

Own the conversation before your competitors.

See where your brand appears, where it doesn’t, and exactly how to win more visibility across search, AI, local, social, and every channel that matters.

Start your free trial

Compliance is a business imperative

This level of discipline and vigilance matters because manual spam actions rarely arrive at convenient moments. More often than not, undesirable Google scrutiny coincides with critical periods: just before a long-planned commercial expansion, in the run-up to a migration project, ahead of an acquisition, as the peak retail season begins, or shortly before investor reporting deadlines.

This is hardly intentional. It’s simply a matter of unfortunate timing. Google doesn’t align search quality enforcement with business planning calendars. Google cares primarily about user experience. For every website that loses its top position, there is usually another capable of providing users with a similarly compelling experience.

Businesses that ignore unresolved policy exposure often discover the problem the hard way, only after search visibility has collapsed and online sales have followed suit. At that point, recovery becomes a far more prolonged, expensive, and operationally disruptive undertaking than ongoing compliance reviews would have been prior to penalization.

Nevertheless, the work must be done. The one silver lining is that, in many cases, the process proves cathartic. Once the penalty has been resolved and the website’s SEO signals have become more consistent, the removal of legacy issues often allows rankings not merely to recover, but to exceed their previous highs.

Read more at Read More

How AI is merging paid and organic visibility

/0 Comments/in , /by
How AI is merging paid and organic visibility

The idea that AI is killing advertising misses the bigger shift. As AI expands across search, assistants, productivity tools, and transactions, advertising is moving with it.

Ad density may be changing within AI experiences, but advertising opportunities are expanding across a growing number of surfaces.

At the same time, paid and organic are becoming harder to separate. The same AI systems increasingly power ad campaigns, search experiences, and brand visibility across Google’s ecosystem.

That changes how brands should think about visibility.

Paid and organic are no longer separate channels competing for the same click. They are increasingly different ways of influencing the same AI systems, which means the signals shaping organic visibility may also affect paid performance.

The old model: Paid and organic on one finite SERP

Google’s SERP was a finite surface: 10 organic blue links, a few ad slots, and a knowledge panel on the right. The user landed, scanned, and clicked.

Paid and organic teams operated on separate budgets, separate tools, and separate quarterly reports, and rarely talked to each other because manual Google Ads kept the paid specialist busy full time. Titles, descriptions, bids, and campaign structure were all chosen by hand and required constant attention, which is why the organic team had no part in any of it.

DSA changed that for me. It read my organic pages to decide which ads to run, who to show them to, when, at what bid, and what title to use. I controlled the descriptions. The engine decided everything else, and it did it better than I would’ve done manually because it was reading the same signals the organic side was already optimizing for.

When someone at Google in Singapore explained how PMax worked, I thought, “That’s exactly what I was doing.”

PMax took the DSA logic and extended it across every Google surface simultaneously: Search, YouTube, Gmail, Display, Maps, and Shopping, all in one campaign, with the engine making every placement decision from your assets and audience signals.

AI Max brought the same intelligence into Search campaigns, specifically, with Gemini underneath instead of rules. PMax and AI Max run on the same Gemini brain: one focused on Search, the other spread across every surface, applying the same funnel logic to different contexts with different signal layers on top.

And if Gemini’s understanding of your brand is thin, it fills those decisions with whatever it thinks will work, which isn’t necessarily your brand narrative, and you have no direct way to override it. You train it, or you lose control of your own ads.

Be the brand AI recommends.

See where your brand appears in AI search, where competitors are winning, and what it takes to become the answer AI recommends.

See your AI visibility

The new model: Gemini sits inside every surface, and it carries ads with it

Gemini now sits inside every layer of the Google ecosystem: 

  • Discovery (Search, Maps, YouTube, Lens, News, Discover, and Shopping), productivity (Gmail, Docs, Drive, Photos, and Calendar).
  • Distribution (Android, Chrome, Google Play, Pixel, Wear OS, Google TV, and Nest).
  • Transaction (Google Pay, Wallet, Flights, Hotels, and Travel).
  • Assistive surfaces themselves (AI Mode, AI Overviews, Assistant, NotebookLM, and the Gemini app). 

That’s how many connected consumers spend most of their workday, and most of those surfaces either carry ads now or have the infrastructure to start carrying them.

Microsoft Advertising sits inside Copilot across Bing, Edge, Windows Consumer, Office Consumer, Teams Free, and GitHub. 

OpenAI Ads launched in February for logged-in users on Free and Go tiers in the U.S., placing ads below ChatGPT responses and clearly labeling them as sponsored. By May, OpenAI had opened a self-serve Ads Manager and was expanding internationally.

The ads layer travels with the engine, the engine is everywhere, and ads therefore have the potential to be everywhere. Most brands still treat paid as a separate channel run by a separate team on a separate dashboard, which is a search-era inheritance that was never ideal but now needs to be dropped. 

Performance Max already runs the auction across YouTube, Display, Search, Discover, Gmail, and Maps as one campaign type. Search is one surface among many, and the “ads are dying in AI search” narrative is measuring the wrong thing. It sees ad slots compress inside the assistive interface while ignoring that the surface base has multiplied by an order of magnitude.

Ad density follows the delegation the user has made to the machine

The dominant narrative in 2026 is that ads are dying because AI is replacing search, and ads inside AI are a problem nobody has fully solved yet. That’s partially correct: Ad density per session drops as AI takes more control, and nobody – including Google – has yet figured out how to insert ads into the AI response itself without killing the experience that makes the AI valuable in the first place.

But this is the part the analysis gets wrong: This doesn’t add up to fewer ads overall.

Search ads are Google’s goose with the golden egg, and the goose may be slowing down — though nobody outside Google actually knows, because Google doesn’t break out search ad revenue from YouTube, Display, and the rest. That ambiguity is doing a lot of work.

What we do know is that total ad revenue has kept growing even as AI has taken over more of the search experience, which proves the flock is already working.

Kodak invented the digital camera and then buried it to protect film-processing revenue, and we know how that ended. Google appears to be doing what Kodak didn’t: building the replacement while the original is still profitable.

Every surface Gemini sits inside is a new bird in the flock, each laying a smaller egg that grows over time, and when Google finally cracks ads inside the AI response itself, that’s one more goose. The surface base has expanded faster than density has dropped, and the ad-density problem in Search and AI is temporary.

The more the user delegates decisions to the machine, the less room the machine has to surface a paid option. Search keeps the user in charge, so the engine surfaces ads the user might pick. Assistive narrows the options, so a sponsored slot still has a chance. Agentic executes the decision, so the ad has nobody to persuade. Ad density follows that delegation, mode by mode, with AI deciding which brands win at each mode.

Ad density follows the delegation the user makes to the machine
Ad density follows the delegation the user makes to the machine.

Google is running two moves at once, and it seems most people have noticed only the first one. Gemini is taking over the recommendation, targeting, and auction logic on surfaces that have carried ads for years. And Google is adding ads to surfaces where they were previously absent, with AI Overviews now eligible for ads above, below, and within the answer, and AI Mode testing conversational ad formats.

The first move is AI taking over the existing ad business. The second is the ad business expanding into surfaces it never occupied. The net effect is more AI-driven ads across more of the stack than ever before.

The freemium system still works, but the ad is becoming part of the surface

The monetization model that works at consumer internet scale is simple: pay with money, or pay with attention.

  • YouTube is Google’s clearest example — and proof that it works: free with ads, paid without, and the vast majority of users have always chosen ads. 
  • Gmail draws the same line: Where the user pays directly, Google doesn’t insert ads. Where the user pays with attention, Google monetizes it.

I learned about freemium the hard way. When our children’s media company, Boowa & Kwala, survived the dot-com crash, we added a paid tier that removed the ads. Out of a million unique visitors a month, a few hundred paid. Almost nobody chose to pay. 

The freemium contract — free access in exchange for ads — is the deal they actively prefer, and the numbers prove it. And for ad-driven businesses, pure volume makes the money. In Big Tech, Google has the clear advantage.

  • ChatGPT is already running ads on free tiers. 
  • Gemini is ad-free without login, but that’s a launch state, not a permanent model. 
  • Perplexity is blocking users instead of monetizing them, which is a different bet on the same problem — and a bet with a limited runway. 

Every AI surface is in the process of landing on the same answer because there is no other answer.

What changes is how the ads appear. The classic SERP ad was clearly labeled and set off in a colored panel. The Gemini recommendation that surfaces a product inside a Gmail context, the Copilot suggestion that names a vendor inside a Word document, and the agent that picks a supplier on the user’s behalf are something else entirely. 

The ad becomes ambient. It dissolves into the surface, and what advertising looks like becomes harder to identify as advertising. Gemini reads context and intent with enough precision that an ad placed in a meeting summary can feel useful rather than disruptive, which is a risk profile Google’s rules-based systems could never have accepted.

At Boowa & Kwala, when we scaled free ad-supported views from 100 million to 1 billion, revenue multiplied by roughly two, and costs rose by around 20%. Surface (a.k.a. pageviews) multiplied tenfold, revenue doubled, costs grew by a fifth, and we went from profitable to significantly more profitable. 

The aim was never to push revenue up at the same rate as surface expansion. It was to keep expanding the surface, knowing the incremental delivery cost was negligible. 

Google’s ratios at planetary scale differ from ours, but the structural shape almost certainly doesn’t: surface expansion plus near-zero incremental cost equals profit growth, regardless of whether revenue per surface keeps pace.

Get the newsletter search marketers rely on.

See terms.

Cohort, intent, and profit drive both paid and organic

PMax, AI Max, AI Overviews, AI Mode — Gemini is driving all of them. The AI optimizing your paid campaigns is the same AI evaluating your organic content, reading the same user, in the same moment, with the same intent.

The engine reads three signals: 

  • Cohort.
  • Intent.
  • Profit. 

In paid, you declare all three explicitly when you structure your campaigns. In organic, the engine infers all three from behavior: clicks, dwell time, and return-to-search serve as proxies for the profit signal that is missing there. Google denied using behavioral signals for years. Its own court case documentation told a different story.

Which means the organic discipline the whole series has been building — the funnel query pathway, the entity home, and the corroboration stack — has always been pointing at one thing: engineer the page so precisely for the right cohort that the behavioral signal does the same job as a correctly structured PMax campaign. The user lands, stays, converts, and doesn’t go back and research the same thing again. Google reads that behavior and infers your profit tier.

My bet, and I want to be clear it’s a bet rather than a documented fact, is that Gemini can’t serve a paid ad in real time without grounding against current search results because the ad has to match the organic context it’s appearing in. 

If it doesn’t ground, the ad is inconsistent with what the user sees organically, which breaks the experience and loses the click. So the grounding process for paid is the same process as for organic: same knowledge graph, same search index, same LLM. 

That means training Gemini on your brand through organic improves your paid performance through the same mechanism. One training investment, two outputs. I’ll be proven right on this eventually, and this article is the timestamp.

The same AI runs your organic and your paid. Train it once, win twice.
The same AI runs your organic and your paid. Train it once, win twice.

You can’t directly target Gemini in AI surfaces. You can only train it.

Across AI-driven placements, Gemini decides everything: where to show your ad, what to show, how to show it, who to show it to, when, and at what bid. The advertiser feeds it information and sets the parameters, but Gemini makes every decision that matters.

What you’re buying when you spend on Google Ads in 2026 is the right to feed a recommendation system that analyzes your brand on its own terms. The explicit signals you declare in paid — cohort, intent, and profit — are a real advantage over organic, where the engine has to infer all three from behavior. 

But your ability to dominate through pure campaign structure is vastly reduced when Gemini doesn’t understand or trust your brand. The control has shifted: you guide it through signal clarity, not through the settings dashboard, and that guidance works best when your organic foundation is solid.

Use paid to find the combinations that work, build organic pages around them

In a correctly structured PMax or AI Max campaign, you declare cohort, intent, and profit margin explicitly: this audience, this goal, this margin, in the same campaign. You don’t mix a luxury hotel and a budget guesthouse in the same ad group because the cohort is different, the profit margin is different, and handing the engine a mixed signal makes it spend your budget resolving a contradiction you created.

Organic doesn’t let you declare profit directly. The engine infers it from who landed, who stayed, who converted, and who never came back to search for the same thing. That behavioral signal is the only proxy it has for the profit tier, and it’s a thin signal compared to the explicit declaration you make in paid.

The smartest move for any brand running both is to treat them as a single loop. Run paid to find which cohort-intent-profit combinations actually convert. Build the organic pages around those combinations, designed so precisely for the right cohort that the behavior on the page sends the engine the same signal the paid campaign explicitly declared.

The paid shortcut in the funnel

The paid side becomes cheaper because organic pages provide the behavioral confirmation the engine needs. The organic side gets stronger because the paid data tells you exactly which pages to build and for whom, and then feeds the engine the same signal the paid campaign declared explicitly, for free.

Most travel sites serve the same page template to a budget traveler looking for a €30 guesthouse in Bangkok and a wealthy traveler looking for a €3,000 suite at the Peninsula. Same layout, same fields, same photo grid, same review format. 

The engine has to infer which cohort the page serves mostly from behavior because the differentiation of the pages is limited. Build the page for the person rather than the query, and you hand the engine the cohort signal it’s currently having to guess. That’s not a UX decision. That’s your profit margin declaration to an engine that can’t see your margins any other way.

And you win on all three fronts simultaneously. A page built precisely for the right person converts better because it works better for the human.

Better conversion behavior sends cleaner implicit signals to the engine, which improves your organic ranking for that cohort. And cleaner organic signals reduce your paid CPC because the engine has less to guess about. Better pages, more organic, cheaper paid – the same work produces all three.

When Gemini isn’t convinced about you, you pay on both sides simultaneously

The three revenue taxes — the doubt tax, the ghost tax, and the invisibility tax — operate on the organic side. Because the engine powering your organic results is the same one powering your paid placements, you pay all three on both sides simultaneously.

  • The doubt tax: When the engine hedges on basic facts about you organically, it rewrites your paid creative to soften the same claims.
  • The ghost tax: When the engine prefers competitors in organic comparisons, your paid creative gets passed over even when your bid is competitive.
  • The invisibility tax: When the engine doesn’t surface you organically, it doesn’t show your ad either. You’re not in the running.

Paid surfaces carry two additional taxes that don’t exist on the organic side, and one discount you earn when you get it right.

The taxes and discounts in AI-driven paid search

The taxes and discounts in AI-driven paid search include:

  • The mistrust tax: What you pay when the engine’s confidence in your brand is low. A CPC premium because Quality Score penalizes low entity trust, and message distortion because the Gemini Filter rewrites your creative away from your intended positioning. You can’t turn the filter off. The practical answer isn’t constraining it. It’s improving the entity confidence that the engine reads when deciding how to filter.
  • The intent tax: This is self-inflicted. Build an ad group with mixed intent, and you hand the engine a contradiction. Gemini will spend your money figuring out a mess you made. Each ad group should align on cohort, intent, and profit margin — any mix across those three, and Gemini is billing you to resolve the confusion.
  • The confidence discount: This is the blade cutting the other way. Every properly defined ad group is secretly doing two jobs: it buys you an efficient placement today, and it teaches the engine which cohort you serve tomorrow. When the engine trusts you, it stops second-guessing your ads, your CPC drops, and your creative lands cleaner. That’s worth more than any bid adjustment you make.

If AI can’t find you, customers won’t either.

Track your visibility across AI search, uncover missed opportunities, and grow your presence where customers are asking questions.

See your AI visibility

Google has a structural advantage that Microsoft and OpenAI can’t match

Google has all the cards: the model, the surfaces, and the ads platform, all owned and tuned together in absolute harmony. Microsoft has the surfaces but lacks the LLM to drive them at the same level. 

OpenAI has the model and launched a real ads business in February 2026, but lacks the surfaces – no Gmail, no YouTube, no Maps, no Play – and without surfaces, an ads business can’t compound at scale. Only Google has all three working as one system.

Paid and organic are now inseparable. The goose is fading, but Google can afford to let it. They know it rises like a phoenix, and in the meantime, they’ve got the biggest gaggle.

This is the 18th piece in my AI authority series.

Read more at Read More

How AI helped build hreflang XML sitemaps at scale

/0 Comments/in , /by
How AI helped build hreflang XML sitemaps at scale

As AI tool usage has become more common, I’ve seen impressive examples of people building tools to automate complex processes that once required significant manual effort. I’ve also seen teams adopt AI simply because it’s available, often with little practical benefit.

My approach is to focus on AI applications that save time and solve real problems.

Recently, I needed to align the SEO architecture for more than a dozen websites across three separate businesses, eight regional domains, and multiple languages, including three English dialects, Italian, Japanese, Spanish, Thai, French, and Korean.

Historically, mapping thousands of URLs to create cohesive hreflang XML sitemaps would have required specialized software or days of spreadsheet work. Instead, I used Google Gemini to build a custom Python script that handled the heavy lifting.

Here’s how the project evolved from an initial prompt into a highly customized automation tool, and what it taught me about using AI for technical SEO.

Where AI delivers the most value

I use AI primarily for practical, time-saving tasks, including:

  • Generating regex patterns when I need a quick solution without researching syntax from scratch.
  • Creating complex spreadsheet formulas for reporting workflows that rely on manual data exports.
  • Accelerating research and planning for projects that require competitive analysis across multiple business lines.
  • Building custom automation tools for recurring SEO and data-processing tasks.

The hreflang project discussed here falls into that final category.

Be the brand customers find first.

Track, grow, and measure your visibility across Google, AI search, social, local, and every channel that influences buying decisions.

Start your free trial

Mapping hreflang at scale

The challenge was clear: map thousands of URLs across more than a dozen multilingual websites into accurate hreflang XML sitemaps.

Rather than tackling the project manually, I used Google Gemini to help build a custom Python solution.

Here’s how the process unfolded.

Phase 1: Asking for an approach, not just a script

A common pitfall when using generative AI for coding is asking it to sprint before it knows the route. If you simply type, “Write a Python script to create an hreflang sitemap,” you’ll get a generic, fragile piece of code that breaks the moment it encounters real-world data.

Instead, I started by asking for an approach. I explained the scenario: multiple regional domains, organic growth over several years resulting in mismatched URL slugs, translated subfolders, and appended revision years.

Gemini suggested a multi-step, data-driven approach:

  • Crawl the websites to collect live URLs and their metadata.
  • Use Python in Google Colab to process the raw data.
  • Run an exact match cluster first to group identical slugs.
  • Use an advanced semantic AI model (such as SentenceTransformers) to fuzzy match translated pages based on their titles and normalized URLs.

Phase 2: Crawling and data collection

Following the strategy, I used a crawler to spider all the regional websites. The goal was to generate a unified comma-separated values (CSV) file containing the live URLs, status codes, title tags, and H1s. Screaming Frog worked perfectly for this application.

A critical point: Your AI output is only as good as your crawl data (remember the old saying, “garbage in, garbage out”).

An AI script will fail to map an obvious “exact match” if the target URL is a 404 or a 301 redirect in your source data. You must filter your CSV to include only indexable content before feeding it to the script.

Dig deeper: International SEO in 2026: What still works, what no longer does, and why

Get the newsletter search marketers rely on.

See terms.

Phase 3: The Google Colab sandbox

Google Colab provides a free, cloud-based Jupyter notebook environment where you can write, paste, and execute Python code without worrying about local installations or environment variables. You can access it through Google Drive. I found the free version had enough capacity to handle this project.

I uploaded the CSV to Colab, and Gemini provided the initial Python script. The script used a domain-mapping routine to assign language codes, clean the URLs, and generate an XML tree. The initial output was far from perfect.

Phase 4: The iteration (where the real work happens)

If you expect AI to deliver a flawless, edge-case-proof script on the first try, you’ll be disappointed. You’ve probably heard the comparison of AI tools to interns, meaning you need to check their work. That’s very true.

The real value of AI lies in the iteration. As we ran the script, we encountered several unmatched URLs, leaving pages orphaned rather than grouping them with their international counterparts.

Here’s how I iteratively trained the AI to handle the nuances of human-managed websites.

The directory flattening problem

The U.S. site had recently reorganized its blog into topical folders, while the Mexican and Italian sites hadn’t yet been reorganized.

I prompted Gemini with these specific mismatched examples. It responded by adding a URL flattener function to the script, which stripped the topical folders behind the scenes so the translated slugs could align cleanly.

The aggressive semantic trap

To prevent the AI from mixing up different topics, we implemented concept traps. Initially, they were too strict. A UK article about the manufacturing sector wouldn’t match an Italian article because the U.S. title was slightly more generic.

I instructed Gemini to loosen the traps for generic industries while keeping them strictly enforced for critical acronyms (such as “SEO” versus “SEM”). This gave the AI the breathing room it needed to match creative translations.

The translated slug epiphany

The biggest breakthrough came while auditing the Mexican blog orphans. For example, the Spanish URL /detras-de-escenas-historias... is a direct translation of the English /behind-the-scenes-stories... I pointed this out to Gemini.

Instead of forcing me to hard-code hundreds of manual matches, Gemini updated the script to create a “Combined Semantic Signature.” It dynamically translated core operational phrases in the slugs, effectively bridging the language gap for the semantic matching model and connecting dozens of orphaned pages almost instantly.

Dig deeper: Cultural SEO: A practical framework for Spanish markets in AI search

Own the conversation before your competitors.

See where your brand appears, where it doesn’t, and exactly how to win more visibility across search, AI, local, social, and every channel that matters.

Start your free trial

Lessons from building an AI-assisted SEO tool

The project reinforced a simple lesson: AI works best when it’s treated as a collaborator rather than a shortcut.

  • Be the strategist, let AI be the coder: Don’t just demand a final product. Discuss the architecture, edge cases, and logic first. Treat AI like a junior developer that needs clear architectural direction.
  • Provide concrete examples: When a script fails, don’t just say, “It’s broken.” For this project, I provided either exact URLs that failed and the URLs they should have matched with, or groups of URLs with mismatches. AI needs concrete patterns to fix its logic.
  • Embrace the iterative loop: Expect to run the code, identify anomalies, and feed them back into the prompt. Each iteration makes the tool significantly smarter.
  • Leverage Google Colab: You don’t need to be a Python expert to use Python for SEO. Colab bridges the technical gap, allowing you to run complex data science libraries directly in your browser.

By the end of the project, we had a robust, highly customized Python script that could process a massive CSV and generate a cross-referenced hreflang XML sitemap in minutes.

AI isn’t going to replace technical SEOs anytime soon. However, SEOs who know how to collaborate with AI to build custom, scalable, and useful tools will have a significant advantage.

Dig deeper: How AI search defines market relevance beyond hreflang

Read more at Read More

Headline formats and Google Discover: What 3.4 million articles reveal

/0 Comments/in , /by
Google Discover headline formats

You’ve probably seen some version of these three claims:

  • Quote-led headlines outperform plain declarative ones by nearly 29%.
  • Question headlines underperform both, sometimes by 24%.
  • Format drives the result: Rewrite a statement as a quote, or add that magic word, and you should expect a real lift.

We tested all three against 1,674,518 English editorial articles and 1,690,295 French articles from the 1492.vision Discover corpus (November 2025 to May 2026): about 3.4 million editorial articles with at least one capture across our fleet.

They share a deeper flaw than any of their numbers.

All three treat headline format as a cause — a lever you pull to gain visibility. But the data shows, layer after layer, that a format’s measured effect is almost entirely a proxy for something else: which publisher used it, for which audience, and on which Discover surface.

The headline is a symptom of those choices, not an independent driver.

The clearest demonstration is Simpson’s paradox. Once you see it, you find it throughout the dataset.

A note on what we measure

Our metric isn’t clicks from Discover; no third party has that data. It’s hits per article: how often an article appears across the 1492.vision fleet we observe, a proxy for visibility.

The corpus is limited to editorial articles. YouTube and X are excluded because their headlines follow different conventions. We’ll return to both at the end—they sharpen the point more than anything else.

A word on why the volume matters: the entire argument depends on being able to slice 3.4 million articles by publisher, Discover surface, topic, and language while still retaining enough data in each segment for meaningful comparisons. That’s the difference between a number and an insight — and between a real format effect and a statistical mirage.

The number is real, at the wrong altitude

Mean hits by headline format, EN and FR

Pool all publishers together, and a clean gradient emerges: quote-led headlines at the top, statements at the bottom.

Lang Format Articles Mean hits Median vs statement
EN Quote-led 38,044 13.0 4 +37%
EN Quote inside 75,463 11.5 4 +21%
EN Question 53,081 10.2 4 +7%
EN Statement 1,674,518 9.5 3 baseline
FR Quote-led 179,472 52.8 13 +48%
FR Quote inside 223,052 49.9 12 +40%
FR Question 103,117 41.3 11 +16%
FR Statement 1,690,295 35.7 9 baseline

The commonly cited +29% is conservative for pure editorial articles: quote-led headlines show a +37% lift in English and +48% in French. Questions, far from underperforming, also outperform statements (+7% EN, +16% FR).

At this level of aggregation, claim 1 looks understated and claim 2 looks plainly wrong.

This is the level of aggregation where most headline advice is born. Hold onto that +37% figure — the rest of this piece is about what it’s actually measuring.

Hidden variable 1: which publisher

The aggregate can’t answer a crucial objection on its own: the publishers that use quotes aren’t the same publishers that don’t.

Celebrity media, regional dailies, and buzz-driven sites lean heavily on quotes and earn more Discover hits per article regardless of headline format. Pure-play publishers, wire services, and utility-focused sites favor declarative headlines and tend to sit lower.

The raw comparison, then, isn’t quote versus statement. It’s one publisher population versus another.

Simpson’s paradox in one chart

This is a textbook Simpson’s paradox: a strong trend in the aggregate that weakens, disappears, or reverses once you segment by group.

To get anywhere near the effect of headline format itself, the grouping variable has to be the publisher.

So make each publisher its own baseline: compare quote versus statement within the same site, holding audience and topic mix constant.

Across 324 English and 439 French publishers with enough of both formats — at least 50 quote and 200 statement articles each:

Within-publisher: does quote beat statement at the same site?
Lang Publishers Quote wins (median site) Quote wins (mean site) Median within-publisher Δ
EN 324 31.5% 55.9% +3.1%
FR 439 47.6% 57.4% +5.5%

In English, statements outperform quotes at 68% of publishers by the median; quote-led headlines hurt more often than they help. In French, the result is close to a coin flip.

That leaves the underlying format effect at roughly +3% to +5%—about five to nine times smaller than the aggregate figure.

(The mean is higher than the median because a minority of publishers see large gains from quotes. The median is the more reliable measure of the typical publisher.)

Stop here and the lesson sounds like “segment your data.” But the collapse points to something larger.

If three-quarters of a +37% effect was really a publisher effect, the obvious next question is: what else is the headline metric standing in for?

The rest of this article is a tour of those hidden variables. And by this point, the answer to claim 3 is already coming into view: the format itself isn’t the driver.

The same substitution, in reverse: questions

Questions: same Simpson, opposite direction

The conventional advice says questions underperform by roughly 24%. The aggregate view of our data says the opposite: questions outperform statements (+7% EN, +16% FR).

Both conclusions are wrong for the same reason. Question headlines are disproportionately used by high-engagement publishers, which inflates their aggregate performance.

Within publishers, the picture settles.

In English, question headlines show a modest real underperformance (-3.7%), winning at only 29.3% of sites. In French, the effect is essentially neutral (-0.5%), with questions outperforming at 46.2% of sites.

The conventional advice gets the direction roughly right in English and neutral in French, but its usual magnitude is about sixfold too large.

The question mark isn’t the cause. The kind of publisher using it is. Same hidden variable, opposite sign.

The effect won’t even hold still

Monthly within-publisher quote advantage

Even that modest within-publisher effect drifts from month to month.

In English, it peaks at +2.5% and turns negative in March 2026, while statements outperform questions at 55% to 60% of sites each month. In French, it ranges from +3% to +12% — strongest in December and February, weakest in March — with no clear trend.

A genuine causal lever shouldn’t wobble like this. A correlation tied to a shifting content mix should.

Hidden variable 2: Which audience

Top EN publishers, quote vs statement

The +3-5% average hides a sharp, consistent split. In English:

  • Gainers: International general news (BBC +85%, Forbes +46%, CBS News +43%, Boston Globe), Yahoo aggregators, mass-market magazines (Parade, Good Housekeeping), Gizmodo.
  • Losers: Specialist sport (RugbyPass, Planet F1, ThisIsAnfield), entertainment (IMDb, TVInsider, People), and factual-leaning dailies (Standard, Washington Post).
Top FR publishers, quote vs statement

French data follows the same pattern in a different market.

  • Gainers: Regional newspapers (La Dépêche, La Montagne, L’Écho Républicain) and general-interest magazines (Grazia).
  • Losers: Specialist sports outlets (Foot National, le10sport, MadeInFoot), technology publishers (Les Numériques), and service-oriented titles (Journal des Femmes, Femme Actuelle).

The pattern is editorial, not algorithmic. Quotes tend to work where the audience comes for commentary, reaction, and framing, and fail where the audience comes for facts.

A publisher built around “what someone said” benefits from a quoted headline. One built around “what just happened” usually doesn’t.

The convergence between English and French is the giveaway. This isn’t a language effect; it’s a reader-intent effect.

What looks like a headline-format effect is, in this case, an audience effect wearing the clothes of a headline.

Hidden variable 3: Which Discover surface

Discover isn’t a single feed. It’s a collection of pipelines, each selecting articles in different ways:

  • Editorial curation (moonstone, mustntmiss).
  • The main topic-personalization engine (aura).
  • Related-reading context (paginationpanoptic, content).
  • Similarity-based recommendation (relatedcontentruby, userpersonascontent).

First, rule out the obvious alternative explanation. Are quote-led articles simply being routed to higher-value Discover surfaces, making the apparent bonus a placement effect rather than a headline effect?

The data says no.

Comparing where quote and statement articles actually appear, the distributions are nearly identical. In English, the largest differences are small: content.f (+2.2 percentage points), aura.f (-1.9), and moonstone.f (+0.6).

Pipeline mix by format

The bonus isn’t about placement: quotes and statements appear on the same surfaces in the same proportions. It’s about intensity — how each format performs once it’s on a surface. There, the overall +3% to 5% breaks into a wide range: from +22% to -14% in EN and from +25% to -12% in FR.

Quote bonus by pipeline, EN, full picture

Grouped into functional families, the pattern is readable:

Pipeline family EN FR
Editorial curation (moonstone, mustntmiss, astria, news…) +3.4% +9.7%
Related reading / context (paginationpanoptic, content…) +2.0% +6.7%
Trends / freshness (deeptrends, freshvideos…) +4.4% +2.3%
Main personalization (aura) +0.6% +1.8%
Similarity-based recommendation (relatedcontentruby, userpersonas…) -1.6% -1.9%

Quote-led headlines win where multiple headlines compete for attention at once — curation carousels, news clusters, and other surfaces where the title carries a social signal: someone said this. They lose on similarity-based recommendations, where the surface sells continuity (“because you read X, you’ll read Y”) and a quote disrupts the topic-clear promise with an out-of-context citation.

The largest pipeline by volume, Aura, ranks on topic affinity and barely reacts to format at all, with gains of just +0.6% to +1.8%.

Why is the net effect so small?

A single quote-led FR article doesn’t get one number; it gets a blend:

  • +10 to +25% on its curation share (moonstone, mustntmiss, astria)
  • ~0% on its aura share, the largest slice of volume
  • -3% on its relatedcontentruby share (≈ 10% of captures)
  • -2 to -6% on shopping/viewer-related surfaces

Integrate those and you land at +4% to +7% net. The curatorial gains are real but partly offset by recommendation losses, which is why the aggregate is nowhere near +29%. The same format is both an asset and a liability, depending entirely on the surface serving it.

And +4–7% overstates how much the format itself matters because each pipeline’s ranking is a compound of signals unrelated to the title: engagement, scroll depth, topic affinity, E-E-A-T, entities, reading history, location, timing, and prior interactions.

A quote in the headline is, at best, one weak signal competing with all of those. Long before an article reaches a feed, it’s largely swamped by everything else.

Questions by pipeline, same story sharper

Question vs statement bonus, by pipeline

These are within-publisher medians (each publisher against itself), so they aren’t a crude artifact of FR using more questions. The format follows the same pipeline logic as quotes, but in a more polarized form:

  • FR curation leans positive on questions; EN curation leans negative. astria.f, the same pipeline in both languages, runs +9% in FR and -1% in EN; FR mustntmiss.f is +14%, EN moonstone.f is -13%.
  • Similarity-based recommendation penalizes questions everywhere, harder than quotes: relatedcontentruby.f FR -11.5% (306 publishers), EN -6.1% (119); itemitemcollaborativefiltering.f FR -14.5%.
  • aura stays neutral in both (+3.5% FR, -0.6% EN).

Two caveats point in the same direction:

  • A fleet-capture metric can’t distinguish an algorithmic penalty from an audience-eviction effect: readers see a question mark, decide “not now,” and scroll past. The fact that relatedcontentruby — which serves already-engaged readers — penalizes questions this heavily points to a behavioral signal, not just ranking.
  • Within-publisher pairing controls for each publisher against itself, but the median is still computed across a different set of publishers in FR and EN, on partly different surfaces. So “FR rewards questions, EN doesn’t” describes the publishers and topics occupying each cell, not an inherent property of the language or the question mark. It’s another hidden variable mistaken for a format effect.

Hidden variable 4: Which editor, and which judgment

Even the honest +3% to 5% comes with a caveat that outweighs its size. When a publisher writes a headline as a quote, they choose the best available quote for that story. So the within-publisher figure compares the best quote an editor selected with the average of all that publisher’s statements, not the same article written two ways.

It’s the subject-line A/B testing problem: a good alternative beats a bad one, but the average alternative doesn’t. Convert every headline to quote-led and you’d be writing average quotes, so most of the gain would disappear. The +3–5% is an upper bound on a selective practice, not the return from a blanket rule.

That’s the final reason “do it everywhere” fails:

  • Not every article has a quote. A sports result, a press release, a market analysis, a product test: forcing one means fabricating it.
  • The editor-selection bias above: The measured bonus is the best quote chosen, not a property of the format.
  • Recommendation pipelines are long-tail levers. relatedcontentruby and friends are how an article redeploys after its initial peak, the main mechanism for extending Discover lifetime. Optimizing the headline for the curation peak while breaking the promise on these surfaces can net negative.
  • The largest pipeline barely reacts. aura is 11% to 15% of FR captures and 7% to 9% of EN, with a +0.6% to 1.8% quote effect. A universal quote rule optimizes secondary surfaces while ignoring that the biggest one runs on topic affinity.

The clincher: the same format, opposite meaning

YouTube and x.com, quote bonus

We excluded YouTube and X from the main corpus, but their results are the clearest proof of the thesis. The same quote-led format produces opposite effects depending entirely on what the title is trying to do.

Domain Lang Quote articles Statement Mean hits quote Mean hits stmt Δ
YouTube EN 43,476 734,986 11.6 10.2 +14%
YouTube FR 16,509 93,912 59.0 29.1 +103%
x.com EN 34,156 268,175 5.2 4.9 +6%
x.com FR 32,201 114,914 21.4 24.6 -13%

On YouTube, the title is effectively a text thumbnail that has seconds to create curiosity. A quote serves as a content promise — “here’s the line worth hearing” — which helps explain the +103% result in French. On X, the title is the post itself, and a detected quote usually indicates that someone is repeating or responding to another person’s words, diluting the original message. That correlates with a -13% result.

Same characters. Same regex. Opposite outcome. The format didn’t change; the job it was doing did.

(Methodological footnote: a naive audit that folded YouTube into the editorial corpus would inflate the overall quote bonus by 20–30 points, while one that folded in X would dilute it. Any serious headline study has to isolate editorial articles before measuring headline effects.)

The headline was never the variable

Put the layers together. Three-quarters of the +37% raw bonus was explained by publisher differences. What remained split again by audience, then by Discover surface, then by which quote the editor selected, and finally reversed entirely when the title served a different function on another platform. At every step, removing context shrank or flipped the apparent format effect.

There’s no clean residue at the bottom where the headline acts independently. The effect is inseparable from the context that creates it.

That’s not a measurement failure; it’s the finding. We just saw the mechanism. Headline format is one weak signal among many stronger ones, all moving through pipelines that often pull in opposite directions.

The consequence is the point. An article’s visibility is the running score of that entire contest, not the verdict of any headline rule. A number measured across publishers is downstream of everything that travels with the format: who published it, what topic it covers, what the audience expects, the newsroom’s style and habits, and the conventions of the language itself.

So when an aggregate reports “+29% for quotes,” it isn’t isolating the quotation marks. It’s measuring a correlation with that whole bundle of factors and quietly relabeling it as causation.

None of this means aggregate data is the enemy. Everything above comes from aggregate data, just analyzed at the right level.

The trap is narrower: treating a single cosmetic variable, averaged across publishers that don’t belong in the same category, as a causal lever.

The same index that exposes that mistake also reveals the signals that genuinely drive Discover: which topics a publisher wins on, which entities are accelerating, who dominates a given surface, and what’s trending before it peaks. Those signals aren’t cosmetic, and they aren’t drowned out by stronger forces. They’re the underlying demand that headline format only weakly approximates.

The lesson isn’t “ignore the data.” It’s “stop averaging the wrong variable across the wrong population.”

This is why no cross-publisher average, corrected or not, converts into a rule for your site:

  • Visibility isn’t traffic. Two sites can earn identical Discover visibility on the same article and see very different CTRs because their audiences click for different reasons.
  • No two audiences are the same. A quote that reads as insider commentary to a magazine reader may read as vague or irrelevant to someone scanning sports scores.
  • A cross-publisher average of one cosmetic feature is the average of audiences you don’t have. Segment by your audience, your topics, and your surfaces, and it becomes information again.

The only test that answers your question is the one you run on your own site, with your own audience. Know who you’re writing for, then measure them. Slice the data by your audience, your topics, and your surfaces — not by a single number averaged across everyone.

So what about the three claims?

Each is real as a correlation and useless as a cause:

  • “Quotes beat statements by ~29%”: True in aggregate — larger than +29%, in fact — but mostly explained by publisher differences. At the publisher level, the residue is +3% to 5%, and even that compares the best quote an editor selected against the average of all statements, not the format itself.
  • “Questions underperform”: Directionally true in EN, neutral in FR, but the magnitude is about 6x too large. The actual effect is roughly -4% in EN and ~0% in FR.
  • “The format itself is the driver”: The claim the dataset refutes. The same article from the same publisher, mechanically rewritten as a quote, would not gain the aggregate effect.

The honest version, if you want one sentence to keep:

A quote-led headline can earn roughly +3% to 7% additional Discover visibility for audiences that value commentary and framing (general news, magazines, regional press), especially on curation surfaces, and lose for factual audiences (sports, tech, utility) and on similarity-based recommendation surfaces. There is no universal gain from quotation marks; the popular ~+29% figure overstates the format effect by roughly an order of magnitude. The useful question isn’t “Should I use a quote?” but “Who am I writing for, and which Discover surface drives my traffic?” The only place to answer that is with your own site, not anyone else’s average.

Methodology

  • Data and period: 1,674,518 EN and 1,690,295 FR editorial articles with Discover visibility from 1492.vision proprietary data, collected between 2025-11-01 and 2026-05-19. Editorial articles only; excludes ads, videos, AI Overviews, and showcases. Domain exclusions: x.com, twitter.com, m.twitter.com, youtube.com, www.youtube.com, and m.youtube.com (reported separately above).
  • Headline format detection (regex): Quote-led: title starts with a multi-word quoted phrase (“…”, «…», ‘…’, or ‘X…’:). Quote inside: a quoted phrase appears but not at the start. Question: ends with ?. Statement: everything else. Titles under 20 or over 300 characters are excluded. Detection deliberately errs toward false negatives in the quote bucket, biasing against finding a quote effect, so the +3–5% is conservative.
  • Three layers of analysis: (1) Raw aggregate: all publishers pooled, producing +37% / +48%. (2) Within-publisher: quote vs. statement inside each publisher with ≥50 quote and ≥200 statement articles; we report the share of publishers favoring quotes and the median per-publisher Δ. This neutralizes publisher-mix bias. (3) Monthly evolution: the same pairing, recomputed monthly with relaxed thresholds (≥10 quote, ≥40 statement).
  • Pipeline layer: Captures come from 1492.vision proprietary data, with each row representing one capture on a specific pipeline. For each (pipeline, format, publisher), captures per article = pipeline captures ÷ distinct articles. Within-publisher pairing includes publishers with ≥20 quote (or question) and ≥60 statement articles on that pipeline. A pipeline is shown only if ≥5 publishers qualify. Pipeline families are an empirical grouping (editorial curation, related reading, trends, similarity-based recommendation, and main personalization) that reflects how each surface behaves.
  • Metric: A “hit” is one capture of an article on Discover by the 1492.vision device fleet. It is a visibility proxy, not a visit.
  • Known limitations: (1) No traffic data: the metric is Discover visibility, not clicks, so a format could affect CTR independently without appearing here. (2) Regex detection misses edge cases and is biased toward under-counting quotes. (3) Within-publisher effects compare the best quote an editor selected against the average statement, not the counterfactual of making every headline quote-led. (4) Some negative pipelines have small publisher samples (<10); the consistent direction matters more than any individual magnitude.

Read more at Read More

Query Fan-Out: What It Is and How It Affects AI Visibility

/0 Comments/in /by

Your content can rank on the first page of Google and still never be cited or mentioned by LLMs.

This makes sense once you understand query fan-out, a background process AI systems use to build answers.

When someone asks ChatGPT or Perplexity a question, it doesn’t default to the best-ranking page.

Instead, it runs related searches behind the scenes, pulling from the most relevant and reliable sources, regardless of position.

User query

If your brand doesn’t show up in those searches (whether through your own content or third parties), you’re unlikely to make it into the answer.

High rankings don’t hurt, of course.

But in AI search, coverage and retrievability are king.

In this guide, I’ll teach you how to optimize your content strategy for query fan-out to help increase your AI visibility.

You’ll learn:

  • Why LLMs use query fan-out
  • How it behaves differently across major AI platforms
  • Why it changes how you create and structure content
  • A 6-step workflow for earning more citations in AI search

Free template: Our Query Fan-Out Audit Template includes ready-to-use spreadsheets for logging money prompts, sub-queries, and content gaps — plus a checklist to keep you on track. Download it now to follow along.

First, I’ll dive deeper into how query fan-out works.

What Is Query Fan-Out?

Query fan-out is a process AI search systems use to break a single user query into multiple sub-queries to create the most helpful response.

In other words, the AI “fans” the query out into a series of related sub-questions to build a more complete picture of the topic.

How query fan-out works

It then pulls information from multiple sources — editorial sites, Reddit threads, comparison and product pages — and synthesizes it into a single comprehensive answer.

The query fan-out process

AI systems use query fan-out for a few reasons:

  • Confirm information: A single source might be wrong or biased. Running parallel sub-queries allows the system to cross-reference multiple sources and find consensus before committing to an answer.
  • Handle complex, specific queries: When a question has multiple layers, like comparing two products across price, reliability, and long-term value, fan-out breaks it into manageable pieces that the system can research independently.
  • Answer the real question: Someone searching “best toothbrush” probably also wants to know about price, battery life, and durability, even if they didn’t say so. Fan-out anticipates those needs and gathers evidence upfront.

For example, a search for “best toothbrush” might trigger sub-queries like “best electric toothbrushes [year]” and “best toothbrushes for sensitive gums.”

This helps the AI build a more complete and useful answer:

Sub-Query What It Contributes to the AI Response
Best electric toothbrushes Top-rated picks and editorial consensus
Best toothbrushes for sensitive gums Use-case recommendations
Oral-B vs. Philips Sonicare Head-to-head comparison data
Best eco-friendly toothbrushes Value picks and pricing information

The AI then synthesizes those findings into a single answer that covers everything the user might want to know: top picks, price ranges, use-case breakdowns, and comparisons.

In this way, it anticipates the user’s needs, even though the original prompt (best toothbrush) was just two words.

ChatGPT – Best toothbrush

What Query Fan-Out Is NOT

Now that we’ve covered what query fan-out is, let’s clear up a few common misconceptions.

Query fan-out is not:

  • Keyword research: This is the process of finding terms your audience searches for. Query fan-out is something AI systems do automatically, behind the scenes, every time someone asks a question.
  • People Also Ask: PAA is a visible SERP feature that shows users what else they might want to search. Fan-out happens in the background whether you can see it or not.
  • A fixed set of queries: Only 27% of fan-out sub-queries remain consistent across repeated searches, according to a SurferSEO study. Sub-queries vary by phrasing, user context, and platform.

Further reading: AI SEO Myths, Debunked: A No-BS Guide for Marketers


Why Query Fan-Out Matters for AI Visibility

Understanding what query fan-out is only gets you so far. The real question is: What does it mean for your content strategy?

Here are four shifts that should make you rethink how you approach content.

You Don’t Need Top Rankings to Get AI Citations

Top rankings don’t automatically translate to AI citations.

When AI breaks a query into sub-queries, it pulls the most relevant and complete source for each one, regardless of where it ranks.

ChatGPT cites pages in position 21+ almost 90% of the time, according to a Semrush study.

Perplexity and Google show the same pattern.

Ranking Positions of LLM-Cited Search Results

AI Retrieves Passages, Not Pages

Rather than directing users to a page, AI systems scan your content and synthesize the exact passage that resolves a query.

This means that the earlier you answer a question, the better your chances of being extracted.

The data backs this up.

44.2% of citations in ChatGPT responses come from the first 30% of a page, while 31.1% come from the middle, and 24.7% from the final third, according to growth advisor Kevin Indig’s analysis of 1.2 million ChatGPT responses.

ChatGPT – Citations from intros

You’re Competing Across a Whole Topic, Not Individual Keywords

SEO often revolves around individual keywords. Query fan-out revolves around comprehensive coverage.

That’s why broad, well-connected coverage across a topic (think pillar pages and topic clusters) can help you earn more AI visibility.

Topic clusters

Pro tip: Pages that rank for fan-out queries (not just the main query) are 161% more likely to get cited, according to a SurferSEO AI Overviews study.

Query Fan-Out Collapses the Buying Journey

We were taught that buyers move linearly — awareness, consideration, decision — and have long optimized content for each stage.

The Marketing Funnel

With AI, those stages collapse into one.

A single high-intent question triggers the system to fan out.

It pulls awareness-level context, consideration-level comparisons, and decision-level specifics into one answer.

The entire buying journey can now happen in a single interaction. So your content needs to work across the full funnel, not just the stage you’re targeting.

Pro tip: Want to work through these steps as you read? Our free Query Fan-Out Audit Template has spreadsheets for tracking your money prompts, sub-queries, intent buckets, and content gaps — plus a checklist to keep the full workflow on track.

The Query Fan-Out Workflow: 6 Steps to Earn More AI Citations

This six-step workflow shows you how to earn more AI citations by identifying and targeting high-impact sub-queries.

It’s repeatable, so you can follow these steps for every topic that matters to your business.

Note: Each AI platform handles fan-out differently, from the number of sub-queries it runs to how it cites sources. We cover the platform differences in depth after the workflow.

Step 1: Find Your Money Prompts

Money prompts are the conversational phrases or questions your ideal customer would ask an AI tool when trying to solve the problem your product or service addresses.

Money prompts are:

  • Typically long-tail and highly specific
  • Tied to a real use case or constraint
  • Close to a decision, not just browsing

Think of money prompts as the AI SEO equivalent of money keywords: high-commercial-intent keywords designed to drive sales.

For example, “noise-canceling headphones ” is a keyword.

“What noise-canceling headphones are best for working from home with kids around, and cost under $300?” is a money prompt.

Noise canceling headphones

Look for money prompts where your audience asks questions:

  • Customer support tickets
  • Community forums
  • Sales call transcripts
  • Internal chat logs
  • Google Search Console queries

For example, when I searched for noise-canceling headphones on Reddit, I found multiple money prompts in real users’ posts.

Like this one that asks for the best noise-canceling headphones for telehealth:

Reddit – Telehealth noise cancelling headphones

And this one asking for durable headphones that will last longer than 2 years:

Reddit – Durable noise cancelling headphones

Forums and transcripts are a good starting point. But you’ll need a dedicated tool to find money prompts using real AI search data.

Semrush’s AI Visibility Toolkit tells you exactly what users type into AI tools, along with the AI’s response.

To show you how it works, I’ll use Bose, a well-known headphone brand, as an example.

Note: I’ll be using Semrush to show you how to complete the query fan-out workflow. If you don’t have a subscription, sign up for a free trial of Semrush One, which includes the AI Visibility Toolkit and Semrush Pro.


First, I searched Bose’s domain in the Visibility Overview tool.

The “Topics & Sources” report revealed over 123.7K prompts where the brand already appears in AI answers.

Visibility Overview – Bose – Prompts

Filtering by “noise canceling” let me dig deeper into topic-specific money prompts like “noise-canceling headphones for sensory issues.”

Visibility Overview – Bose – Prompts – Noise canceling

Clicking the prompt provides a full breakdown: the AI’s response, every brand mentioned alongside yours, and the exact sources it cited.

Visibility Overview – Bose – Prompt details

Follow the same process for your own domain.

These prompts are your highest-priority money prompts — your audience is already searching them, and AI is already answering them.

Don’t have AI visibility yet? Use the Prompt Research tool.

Enter a broad topic to see the prompts that generate the most AI results in your industry.

Prompt Research – Noise canceling headphones

As you find relevant prompts, add them to your spreadsheet.

Even a few money prompts give you enough to work with for the next step.

Fan-Out Audit Template – Money Prompts

Step 2: Generate Your Fan-Out Set

There are two ways to generate fan-out sets: manually or with a dedicated fan-out tool.

The manual approach is free and helps you understand how fan-out behaves, while tools are faster and better suited to working at scale.

I’ll start with the manual method.

Paste this prompt template into any AI platform to get a fan-out set:

Expand this question into the sub-queries an AI system might search to answer it: [your money prompt].

When I ran my Reddit money prompt through ChatGPT, it returned sub-queries grouped into categories:

  • “Core Product Category”
  • “Durability & Longevity”
  • “Battery & Hardware Lifespan”
  • “Reliability & Failure Rates”

ChatGPT – Money prompt

Each category is a potential content gap you’ll address in Step 4.

Run your money prompt through multiple AI tools to get a more complete picture, since each platform tends to expand prompts differently.

Pro tip: Manual research is a solid starting point, but outputs can contain inaccuracies or hallucinations. A dedicated fan-out tool simulates how different AI platforms expand your query and returns an organized list of sub-queries you can act on immediately.

For a faster option, Backlinko’s free ChatGPT Query Fan-Out Tool is worth trying.

Install the Chrome extension, open ChatGPT, and ask your money prompt. The extension captures the response in real time and breaks down every sub-query ChatGPT ran behind the scenes.

When I ran a prompt through it, the panel showed:

  • Each sub-query the model generated
  • The metadata behind the response, including model version
  • Every URL cited, categorized by type: sources, products, images, and news

As you gather sub-queries, assign a query type to each — this tells you what kind of content you’ll need to create in the next step.

Use these definitions to categorize them.

Query Type What It Means
Reformulation A reworded version of the original prompt
Comparative Weighs two or more options against each other
Implicit Addresses a need the user didn’t explicitly state
Personalized Tailored to a specific situation, constraint, or preference
Entity expansion Drills into a specific brand, product, or person mentioned
Related A connected topic the AI anticipates the user might want next

Step 3: Bucket Sub-Queries by Intent Type

Bucketing by intent tells you what types of content to create and the ideal format for each.

To categorize a sub-query, answer this question: What does the person actually want to do after getting an answer?

Consider an example from the noise-canceling headphones query fan-out set: “Sony vs Bose Noise Canceling Headphones.”

Someone asking this is weighing two specific products against each other, so it’s a “comparison” query.

Fan-Out Audit Template – Intent Buckets

The right format for this query is a head-to-head comparison page or table, not a general buying guide or listicle.

The intent isn’t always this obvious, and some sub-queries may fit more than one bucket.

When that happens, place it where the strongest intent lies.

Here’s a general guide to the main intent buckets and what each one calls for:

Bucket Description Example Sub-Query Content Format
Definitions / Basics What is X? How does X work? “how do noise canceling headphones work” Explainer article, glossary section
Comparisons / Alternatives X vs Y, alternatives to X “apple airpods max vs sony wh 1000xm4” Comparison page, head-to-head section
Best for X / Recommendations Best option for a specific use case “best noise canceling headphones for working from home” Listicle, buying guide
Problems / Troubleshooting How to fix X, why does X happen “how to get rid of background noise in audio” How-to guide, FAQ section
Pricing / Value How much does X cost, is X worth it “are there any good wireless headphones with noise cancellation under $150?” Pricing page, value comparison section
Social Proof / Discussions Reviews, Reddit opinions, user experience “best earbuds for calls in noisy environment reddit” Review roundup, user feedback section

Step 4: Audit Your Existing Content for Gaps

Once you’ve bucketed your sub-queries by intent and format, check which ones your site already covers and which ones it doesn’t (aka content gaps).

Start by searching your own site.

Type “site:yourdomain.com [sub-query topic]” into Google.

For example, running “site:bose.com noise canceling headphones” surfaces all their pages on that topic.

Google SERP – Bose – Noise canceling headphones

From here, evaluate each page against the sub-query it should cover:

  • Coverage: Does it directly answer the sub-query, or just mention the topic in passing?
  • Format: Is it the right content format for the intent?
  • Self-contained answers: Can the answer stand on its own, without the reader needing to look anywhere else?

Categorize each page by its coverage level:

Coverage Level What It Looks Like What to Do
Not covered No page on your site addresses this sub-query at all Create new content targeting this sub-query directly
Partially covered A page mentions the topic in passing but doesn’t resolve the sub-query directly Add a dedicated section to the existing page that fully answers the sub-query
Fully covered A dedicated section or page answers the sub-query completely and can be extracted and cited by AI without needing surrounding context Monitor for AI citations and update regularly to stay current

For each sub-query, you’ll also want to know which competitors are showing up for your money prompts.

Run your money prompts through AI platforms to gather this information manually. Or refer back to your research from the AI Visibility Toolkit in Step 1.

Click any prompt to see which brands were mentioned and the exact sources the AI cited.

Bose – Prompt details – Brands & Sources

Already showing up alongside competitors? That’s a prompt worth protecting — focus on strengthening your coverage so you stay in the answer.

If competitors are showing up and you’re not, that’s a gap worth closing before they own it.

Fan-Out Audit Template – Content Audit

Step 5: Structure Your Content So AI Can Extract It

Creating the right content is only half the job. The other half is making it easy for AI to find, parse, and use.

Start by filling the gaps you identified in Step 4.

For sub-queries with no coverage, create dedicated pages or sections that target them directly.

For partial coverage, add self-contained answers to existing pages that resolve the sub-query without needing surrounding context.

Then, structure everything so AI can extract it cleanly:

  • Address specific questions directly — lead with the answer, not background context
  • Use content chunking: Break content into focused sections with clear headings, short paragraphs, and bullet points
  • Front-load key information early in the page or section
  • Use clear, precise language, including specific product names, figures, and use-case-specific wording
  • Add FAQ sections

Here’s what this looks like in action.

Bose has over 63.9K mentions across AI platforms in the U.S. alone:

Visibility Overview – Bose

It helps that they’re a household name. But their content is also built to be extracted.

Their product pages front-load specific claims as scannable elements — “24 hours of battery life” and “legendary noise cancelation” — rather than burying them in copy.

Bose – Product features

Key specs are organized into structured comparison tables:

Bose – Product specs

And they build dedicated landing pages for use cases like flying, using descriptive, scenario-specific language.

This matters because AI fans out into use-case-specific sub-queries.

Bose – Noise cancelling headphones for flights

When I searched “best noise-canceling headphones for flight anxiety,” AI Mode recommended Bose, using nearly identical language from Bose’s flight landing page.

Google AI Mode – Noise canceling headphones

When a user’s prompt matches the scenario your page was built for, AI systems may be more likely to pull from it.

This is a clear example of that in action.

You don’t need a complete site overhaul to make this work.

Even restructuring a few high-priority pages to address your fan-out gaps can improve your chances of being extracted and cited.

Step 6: Measure Your Performance in AI Search

Once your content is structured and live, track your performance in LLMs.

Start with the money prompts you identified in Step 1.

For each one, you want to know:

  • Are you showing up? Is your brand mentioned or recommended in the response?
  • Is what it says accurate? Are the claims the AI makes about your brand correct, or is it pulling outdated or wrong information?
  • How do you compare? Which competitors appear in the same response, and how are they positioned relative to you?

If you’re tracking manually, run them through multiple LLMs (in a private or incognito window) and record what you find.

ChatGPT – Bose headphones

But once you’re tracking dozens of sub-queries across platforms, manually tracking gets messy (and time-consuming).

I use Semrush’s Prompt Tracker to automate the process.

It alerts you to changes in mentions for your money prompts, so you don’t have to keep re-running them yourself.

Position Tracking – Keywords

Another helpful tool is the Visibility Overview.

It provides an AI visibility score that tracks how often you’re showing up in AI answers compared to competitors.

Visibility Overview – Bose

The Perception tool tracks sentiment so you know how LLMs describe your brand — and if they mention competitors more favorably.

Perception – Bose – Sentiment

It also breaks down the factors driving that sentiment.

For Bose, “industry-leading noise cancellation” shows up as a strength, while “over-the-ear models not sweatproof” flags a use-case they could address with targeted content.

Perception – Bose – Key sentiment drivers

Tracking should be an ongoing process.

Revisit your money prompts regularly and update your content as new sub-queries emerge or competitors gain ground.

How Query Fan-Out Works Across Different Platforms

How content surfaces in an AI answer depends on several factors:

  • Whether the system searches the live web or draws from its training knowledge
  • How many sub-queries it runs
  • Which sources it favors, and how it cites them

Understanding those patterns helps you make smarter decisions about content structure, format, and where to focus your optimization effort.

Plus, if a competitor outperforms you in a specific LLM, understanding how that platform handles fan-out can help you figure out why.

Platform How Fan-Out Works
ChatGPT Reasons internally, then runs live web searches when a question requires fresh data, comparisons, or current information
Perplexity Combines conversation context with real-time web search
Claude Clarifies intent first; relies mostly on training data
Google AI Overviews Synthesizes Google’s index into condensed, featured-snippet-style summaries
Google AI Mode Breaks complex prompts into multiple searches across Google’s index

Note: Some of the behavior described below is based on how each system describes its own reasoning when prompted. LLMs aren’t always reliable narrators of their own processes, so treat these observations as directional rather than definitive.

ChatGPT

For simple, informational queries, ChatGPT usually responds from its training data without running a live search.

ChatGPT – Compound interest

But that changes when the question requires fresh information, comparisons, or real-world data.

When I asked which car I should buy (Toyota vs. Honda) in Thinking mode, ChatGPT spent about 22 seconds reasoning through the question.

Then, it produced an answer drawn from 41 cited sources

ChatGPT – Toyota vs Honda

That’s query fan-out in action: one prompt, varied sources, and multiple sub-queries running behind the scenes.

By default, you can’t see the sub-queries ChatGPT runs. But I’ll show you how to find them (don’t worry — it’s easier than it looks).

Note: This DevTools method only works in the web version of ChatGPT. You can’t access sub-query data on mobile or in the desktop app.

First, search a money prompt in ChatGPT.

Then, look at your browser’s address bar and copy the slug that appears after chatgpt.com/c/ — that’s the unique ID for your conversation

ChatGPT – URL

Next, right-click anywhere on the page and select “Inspect.”

ChatGPT – Inspect

A developer panel will open on the side of your screen:

  • Click “Network” at the top of that panel
  • Paste the slug you copied into the filter bar
  • Refresh the page

Click on the fetch version of the slug (here, it’s the second option under the Name column).

Chrome DevTools – Network

Then, open the Response tab.

Chrome DevTools – Network – Response

Once it loads, press Ctrl+F (or Cmd+F on Mac) and search for the word “queries.”

Chrome DevTools – Network – Response – Queries

What appears is the exact set of internal searches ChatGPT ran before producing its answer.

For the Toyota vs Honda prompt, ChatGPT generated queries around:

  • Vehicle specifications
  • Fuel economy
  • Reliability
  • Safety ratings
  • Long-term ownership costs

Once you have the sub-queries, cross-reference them against your content.

Are you targeting each one? Do your pages use the same language ChatGPT is searching for — “long-term ownership costs” rather than just “value”?

ChatGPT often pulls from third-party sources like Reddit threads, review sites, and comparison pages.

So topical authority matters here — not just what’s on your site, but whether your brand shows up across the sources ChatGPT is likely to retrieve.

Perplexity

Perplexity runs two types of fan-out simultaneously:

  1. Internal fan-out — scans your prior conversation history for relevant context
  2. External fan-out — searches the external web for relevant information

The final answer draws on both layers, which means your content needs to work for a range of user situations, not just one.

For the Toyota vs. Honda question, Perplexity’s first batch of sub-queries had nothing to do with the cars.

Perplexity – Toyota vs Honda

Instead, it checked whether I’d previously mentioned anything that could shape its recommendation.

Perplexity – Toyota vs Honda – Subqueries

Like budget constraints, driving habits, or past questions about either brand.

Perplexity – Toyota vs Honda – Subqueries – Details

Only after that internal scan did it launch external searches about reliability, ownership cost, and safety ratings.

What this means for your content: Perplexity may pair your page with context you can’t predict: a user’s past questions, constraints, or preferences.

Your content needs to be specific and self-contained enough to remain accurate and useful no matter the surrounding context.

Claude

Claude takes a different approach.

Rather than immediately running sub-queries, it asks clarifying questions first. Then, it generates a response tailored to your answers.

When I asked the Toyota vs. Honda question, Claude presented a preference widget before producing an answer.

Claude – Toyota vs Honda

Once I responded, it generated a recommendation tailored to my priorities.

Claude – Toyota vs Honda – Answer

Because it clarifies intent before searching, Claude tends to generate fewer, more targeted fan-out sub-queries than other platforms.

The implication for your content: Answer specific, well-defined use cases directly rather than trying to cover every angle on a single page.

Google AI Overviews and AI Mode

AI Overviews appear as concise, AI-generated summaries with sources listed in a clickable sidebar.

Google SERP – Toyota vs Honda – AI Overview

They work by synthesizing Google’s existing web index into a tighter, more contained summary.

AI Mode, by contrast, is a dedicated conversational search tab designed for complex, multi‑part questions.

Google AI Mode – Toyota vs Honda

Like AI Overviews, it draws on Google’s index to generate answers, but it offers more interaction and depth.

Neither platform exposes the sub-queries it runs.

But SEOs have found a way to extract Google’s fan-outs using Screaming Frog configured with a Gemini API. Watch Dan Hinckley’s tutorial for a full walkthrough.

For both, the optimization focus is the same: Front-load your answers, use descriptive subheadings, and structure content so individual passages stand on their own.

AI Search Runs on Query Fan-Out — Your Content Strategy Should Too

High rankings alone won’t earn AI mentions.

The brands showing up are the ones covering the questions their audience is actually asking and making that content easy for AI to extract and cite.

You’ve got the query fan-out framework. Now it’s about execution.

Start with one money prompt, map the sub-queries, and audit where your content stands.

Then work through the gaps, one topic at a time.

Next, dive deeper into how to get your brand seen and trusted across AI platforms with our AI search strategy guide.

The post Query Fan-Out: What It Is and How It Affects AI Visibility appeared first on Backlinko.

Read more at Read More

Google zero-click searches hit 68% in early 2026: Study

/0 Comments/in , /by
Google zero

Google searches ended without a click 68.01% of the time in the U.S. during the first four months of 2026, according to new SparkToro research based on Similarweb clickstream data. That’s up from 60.45% in 2024, a 7.56-point increase in two years.

Fewer searches result in clicks. The share of searches generating at least one click fell 9.51 percentage points between 2024 and 2026 (a 22.9% decline), according to SparkToro. This includes clicks to organic results, paid ads, and Google-owned properties such as Maps and YouTube, but excludes follow-up searches within Google.

  • Over the same period, the share of searches that led to another Google search rose 7.2 percentage points.
  • This trend reflects Google’s growing ability to answer questions directly in search results while encouraging users to refine or continue their searches within Google, according to SparkToro.

AI Overviews and zero click. SparkToro believes AI Overviews are likely contributing to the increase in zero-click searches, though the study doesn’t isolate the extent to which the overall rise between 2024 and 2026 can be attributed specifically to AI Overviews.

  • AI Overviews now appear on more than 20% of Google searches, according to the research. When they do, click-through rates drop by nearly 60%.

AI Mode and zero click. It appears to have played only a limited role during the January to April study period. SparkToro found that just 0.34% of searches transitioned into AI Mode during that time.

  • However, Google said at I/O 2026 that AI Mode had surpassed 1 billion monthly users and that query volume was more than doubling each quarter, suggesting its impact on search behavior could grow significantly.

Zero click history. SparkToro has tracked zero-click search behavior for years, though its underlying data sources have changed over time. Because the studies rely on different providers, panels, and methodologies, long-term comparisons are not directly equivalent. Still, the available data consistently points to a rise in zero-click behavior over time, according to SparkToro.

Why we care. The findings suggest Google is increasingly satisfying user needs without sending users to external websites. However, you should interpret direct comparisons across years cautiously because SparkToro’s historical analyses rely on different clickstream data providers and panels.

SEO still matters, but… SEO alone may be insufficient for many publishers seeking to regain historical levels of Google-referred traffic. SparkToro co-founder Rand Fishkin recommended investing in brand awareness and influence on the platforms where your audience already spends time, regardless of whether those efforts drive direct website visits.

  • Some categories continue to benefit significantly from SEO, including branded searches, local business queries, and high-intent transactional searches, Fishkin said.

About the data. The study used Similarweb desktop and mobile web panel data covering U.S. Google searches from January through April 2026. SparkToro assumed that two-thirds of searches occurred on mobile devices and one-third on desktops. The analysis excludes searches conducted in Google’s mobile search app, where SparkToro said zero-click behavior may be even higher.

The study. In 2026, Less than One Third of Google Searches Still Send a Click

Read more at Read More