Tag Archive for: Trends

Many budget allocation strategies assume that every channel follows the same pattern: the first dollar is the most productive, and each additional dollar yields a slightly lower return.

The charts below show what that pattern looks like.

The log shape means that the first dollar is the most productive, and each subsequent dollar is worth a little less. When every channel looks like that, the game plan is to spread the budget to as many channels as possible and equalize the marginal CPAs to maximize profit.

But not every channel looks like that. Some have a warm-up region where the early spend is the least efficient, not the most. On those channels, the logic above breaks, and so does the “test small, scale the winners” playbook that most of the industry runs on autopilot. 

The difference comes down to one question about the channel: Is the response curve C-shaped or S-shaped?

The answer can change how you approach channel testing and channel measurement, including any MMM analysis. Moreover, Google has been incorporating more S-shaped campaign types, and after its Google Marketing Live announcements, this trend seems set to continue.

The two shapes — and the only part that matters

The response curve plots output (conversions, revenue) against input (spend). This generally results in two types of curves in marketing.

• C-shaped (concave): Diminishing returns from the very first dollar. A log or power curve. Picture the top-left quarter of a circle: steep at the start, flattening as you go.
• S-shaped (sigmoid): A slow, inefficient start, then an inflection point where it gets steep, followed by a flattening into saturation. A logistic curve.

The response curve itself isn’t what you allocate against. You allocate against the marginal curve, the derivative, which answers the question: “What did the next dollar buy me?” That’s where the shapes diverge in a way that matters.

• For a C-curve, marginal return is highest at the first dollar and falls in only one direction. Marginal CPA rises from the first dollar onward. If conversions are a*ln(s), marginal conversions per dollar are a/s, so marginal CPA is s/a, climbing in a straight line as you scale. There’s no warm-up. The cheapest conversion you’ll ever buy is the first one.
• For an S-curve, marginal return starts low, rises to a peak at the inflection point, then falls. Marginal CPA is U-shaped. It’s expensive at the start, bottoms out around the inflection point, then climbs into saturation.

That region of increasing marginal returns is the whole story. It’s the difference between a channel where small budgets are productive and one where they are wasted.

See exactly how your competitors win.

Uncover the keywords, ads, landing pages, and strategies driving your competitors’ paid search success—and find your next opportunity to outperform them.

Analyze your competitors

How this looks in a marketing campaign

Say your CPA goal is $50. Here is an S-shaped channel, modeled as Conversions = 1000 / (1 + e^(-0.25(s – 20))), with spend in the thousands and the inflection at $20,000/month:

Run the $10,000 test that a sane person runs before committing real budget. Average CPA comes back at $132, marginal around $94. If those two metrics are all you look at, you conclude that this channel can’t hit $50, so let’s kill it.

That verdict is wrong. At $20,000 to $25,000, the channel is running at an average of $32 to $40, and the marginal dollar in the $15,000 to $25,000 band costs $18. That’s not “barely viable.” In that band, it’s the best marginal buy you have. The small test fell within the warm-up and reversed the conclusion.

In a C-shaped channel, the small test would have shown you the best the channel can do. On an S-shaped channel, it shows you the worst.

This is the trap. The standard playbook is “test small, scale what works.” On S-curves, small tests systematically condemn channels that would’ve worked at scale because the test is structurally stuck in the inefficient region.

---

The allocation logic, restated

C-shaped channels, go wide

The optimization is convex. There’s one global optimum, the equimarginal rule from the marginal-CPA post applies cleanly, and the solution is usually interior, meaning lots of channels get funded.

Even a small allocation is productive because the first dollar is the best dollar. Run many channels lean, reallocate continuously at the margin, and pull back the instant marginal CPA crosses your goal.

S-shaped channels, go deep or skip

The optimization is non-convex. A small allocation can be strictly worse than zero because below the inflection your marginal return sits under your target, and you’ve sunk money to get nowhere.

The decision isn’t “how much.” It is binary: commit past the threshold, or don’t fund it at all. There’s a real minimum viable budget, and it’s often above normal test budgets. You can’t sprinkle an S-curve and expect efficiency, and you can’t evaluate one on an underfunded test.

Those two rules can look like they fight each other, but that’s only true to a certain point. Past the inflection, an S-curve is concave, so the equimarginal rule governs it exactly as it governs a true C. The S-specific instruction — commit a block instead of sprinkling — is only about the trip from zero to past the inflection.

Shape is therefore mostly a launch-and-evaluation problem. Getting a new prospecting channel into its efficient range requires a committed block and patience with ugly early numbers. Once it clears the inflection, you manage it at the margin like everything else, right up until you consider cutting it hard, where shape matters again because the downside is a cliff, not a ramp.

This is the part that’s genuinely counterintuitive, and it echoes the original marginal-return point: The right move isn’t always the one that looks most efficient at a small scale.

Which channels are which?

The historical default was concave. Simon and Arndt reviewed more than 100 studies and concluded that advertising follows the law of diminishing returns, a concave response. 

The dissent came later: Vakratsas, Feinberg, Bass, and Kalyanaram found that threshold effects do exist and that response is not necessarily globally concave. Their explanation for why thresholds were so hard to find is the useful part. Mature accounts already operate inside the effective range, so the warm-up never shows up in the data, and most studies fit a concave model (the double-log) that can’t reject an S-curve even when one is present.

The platform shift has made the threshold visible again. Here is a fuller map, ordered roughly from C to S. The shape column is an inference from how each system targets and learns, not a measured constant, and the right shape for your account still has to be measured.

Two rows do most of the work.

AI Max is the live example of a channel migrating from C toward S. Swapping explicit keywords for broad and keywordless matching means it needs conversion volume to learn which queries convert, so below a data threshold, it explores badly.

The mixed independent results fit that: Google reports about 14% more conversions on average and up to 27% for exact-match-heavy campaigns, while independent testing reports 84% of advertisers seeing neutral or negative results. Much of that spread is accounts that turned it on without the conversion volume to clear the learning region.

Performance Max is the trap, because its curve is a composite. It blends a harvesting layer (branded, retargeting, Shopping against existing intent) with a prospecting layer (keywordless expansion across surfaces). The harvesting layer is a cheap C that pays off on the first dollar. The prospecting layer is the S underneath.

Blended, the early efficiency looks great, because you are mostly skimming demand you already had, and the average hides the prospecting warm-up entirely. That is also why the platform is glad to optimize it for you: the blend flatters the headline number. You can’t read PMax or run the shape analysis on it until you split the harvesting from the prospecting.

The throughline runs in two layers. Rules-based auctions capture the best inventory first, which yields concavity; machine-learning systems must be fed before they are efficient, which introduces a threshold. Underneath both, harvesting existing demand is concave and mostly non-incremental, while creating new demand is the S-shaped part where the real growth and the real warm-up cost both sit.

Average versus marginal: total over spend, or the slope where you stand.

What you allocate against is marginal incremental return, the slope of the incremental curve at your operating point. A holdout fixes the first axis only. Time-sliced marginal CPA on attributed data fixes the second only. A multi-cell scaling test gets both, at a cost. 

MMM (method 1) estimates the whole curve from aggregate data and sidesteps click attribution entirely, but pays in identifiability and modeling assumptions instead. Most arguments about ‘what is working’ are two people standing on different axes.

There are two major cautions, and I would flag both as genuinely unsettled rather than settled facts. 

• Separating a true S-curve from “concave with a high half-saturation point” is hard, because a concave model will fit S-shaped data well enough to hide the inflection (this is the Vakratsas point, and it applies to your own dashboards as much as to academic studies). 
• The learning phase may be a one-time fixed cost to train the model rather than a permanent feature of the steady-state curve. If it is transient, the channel may behave concavely at the margin once it is trained, and the S you measured was a startup artifact. The truth is probably a mix: a one-time training cost, plus an ongoing minimum-volume requirement to stay efficient. Treat every shape call as provisional and re-check it.

One more failure mode, and this one is not unsettled science but a matter of where you are standing on the curve. An S only looks like an S if your data spans the inflection. 

Above the inflection, an S is concave, mathematically identical to a C. Look at only the $20,000-and-up rows of the table above: marginal CPA rises monotonically from $18, a textbook C-curve, and the convex warm-up is invisible because you are no longer operating in it. 

Established accounts usually sit past the inflection, which is exactly why Vakratsas found thresholds so hard to detect, and why you can run an S-shaped channel for years, correctly, while believing it is concave. The tell arrives the day you cut hard and fall off the inflection instead of easing down a slope.

When to go wide and when to go deep

The marginal-return post told you to equalize marginal CPAs across the program. That rule is still correct, but the shape of the curve tells you how you’re allowed to get there. 

• On C-shaped channels, you can get there by sprinkling, because every dollar is productive and breadth is the natural answer. 
• On S-shaped channels, you have to commit a block of budget past the inflection before the channel earns its place, and then concentrate rather than spread.

Lay the harvest-versus-create cut on top. Harvesting channels (branded, retargeting, non-brand search) are your C-curves: fund the first dollars, then cap them early, because they saturate fast and most of the tail isn’t incremental, no matter how strong the attributed ROAS looks. 

Prospecting channels (Meta, YouTube, LinkedIn, the expansion half of PMax) are your S-curves and your only real source of incremental growth: commit past the warm-up or don’t start, and judge them on incremental lift rather than attributed CPA, or you’ll kill the thing that was working.

Classic search rewards going wide. PMax, AI Max, and Meta prospecting reward going deep on fewer bets and giving each enough volume to clear the warm-up. Run an S-curve like a C-curve and you’ll starve it, read the underfunded result, and kill a channel that would’ve been one of your best.

Read more at Read More

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.

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.

---

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.

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 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 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.

• Part 1, “Rand Fishkin proved AI recommendations are inconsistent, here’s why and how to fix it,” introduced cascading confidence.
• Part 2, “AAO: Why assistive agent optimization is the next evolution of SEO,” named the discipline.
• Part 3, “The AI engine pipeline: 10 gates that decide whether you win the recommendation,” mapped the full pipeline.
• Part 4, “The five infrastructure gates behind crawl, render, and index,” walked through the infrastructure phase.
• Part 5, “5 competitive gates hidden inside ‘rank and display’,” covered the competitive phase.
• Part 6, “The entity home: The page that shapes how search, AI, and users see your brand,” mapped the raw material.
• Part 7, “The push layer returns: Why ‘publish and wait’ is half a strategy,” extended the entry model.
• Part 8, “How AI decides what your content means and why it gets you wrong,” covered annotation.
• Part 9, “Why topical authority isn’t enough for AI search,” opened the competitive phase with topical ownership.
• Part 10, “The funnel flip: Why AI forces a bottom-up acquisition strategy,” named the process.
• Part 11, “The framing gap: Why AI can’t position your brand,” exposed the gap between evidence and recommendation.
• Part 12, “The 10-gate AI search pipeline: Find where your content fails,” showed how to find and repair your weakest gates.
• Part 13, “The delegation boundary: How AI decides which brands win,” mapped how delegation moves across Search, Assistive, and Agent modes.
• Part 14, “The funnel query pathway: A framework for measuring AI visibility,” built the measurement instrument.
• Part 15, “The micro-macro shift: How to measure AI visibility now that precision is gone,” moved measurement from micro precision to macro trend.
• Part 16, “How SEO turns customer success into AI-readable proof,” put SEO inside post-sale operations.
• Part 17, “How AI forms opinions about your brand,” named the discipline that turns evidence into machine legible signal.
• Up next: “The untrained salesforce,” the operational synthesis the whole series has been building toward.

Read more at Read More

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

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.

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:

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

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

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

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).

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).

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.

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

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

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