Search for AI-search statistics and you will find pages listing twenty or thirty numbers, sourced to blogs sourcing other blogs, with no indication of which figures survive contact with their primary sources. Several of the most-cited numbers on those pages are, checkably, wrong. This page is built differently: fewer numbers, each carrying an explicit verdict (verified, refuted, or circulating but unproven) with the reasoning attached. It is the statistics page we wanted to find and couldn't.

How these verdicts were reached.

In July 2026 we ran the most-circulated claims in the AEO/GEO literature through adversarial verification: trace each number to its primary source, read what the source actually measured, and try to refute the claim rather than confirm it. Every verdict is re-checked each quarter against its primary source, most recently in November 2026 (State of AI Search, Issue 1). Verified means the figure is in its primary source and measures what the claim says. It does not mean the question is settled. Verified as dated means it was accurate when measured and should not be read as current. A claim is marked refuted only when a source contradicts it. A claim we cannot confirm is marked unproven. The checked set is maintained, with sourcing, in our standing AI-Search Mythbusting Register, which this page draws on and which gets refreshed as new claims are checked. Where a number below is ours, we say so; where a verdict could change with new evidence, the entry says what would change it.

Start here: the winnable third.

Analysis of ChatGPT's most-cited pages in September 2025 found that only about 32% are the kind of pages content or brand work can realistically influence. The remainder goes to Wikipedia, homepages, app stores, and reference sites that no marketing strategy touches. That is the honest, addressable size of the "AI visibility" opportunity: roughly a third of what the category's sales language implies.

Two things follow. Budgets should be sized against the contestable third, not against total AI answer volume. And within that third, the verified mechanics below say the game rewards specificity and verifiability, which is precisely where a narrow expert practice can beat a large generalist. The ceiling is lower than advertised. Inside it, the evidence favors the specific, checkable expert over the large generalist.

One more calibration before the numbers: being counted in an AI answer is not the same as being represented accurately in it. There are documented cases of an engine blending a source's real content into a quietly wrong composite - the affected business watching a wrong version of its own material served under its own brand name. No statistic below measures that fidelity dimension, because few providers measure it. We track it separately in client work, and any statistics page that ignores it is overstating what "visibility" means.

Dated November 2026: the ~32% figure is a September 2025 measurement and has not been repeated. ChatGPT's source mix shifted during 2026, so treat it as a dated estimate.

Verified: the demand shift.

Verified

8% vs 15% - click-through on traditional results with vs. without an AI summary present

Pew Research Center's direct measurement of real user behavior (March 2025 data, published July 2025). It is observational: visits with an AI summary showed about half the click rate of visits without.

Verified

26% vs 16% - search pages after which the user ended the browsing session, with vs. without an AI summary

Same Pew measurement. It counts people who stopped browsing, which is a different thing from a search that ends without a click.

Verified - one vendor's estimate

~9x - growth in monthly US searches for "generative engine optimization" from October 2024 to January 2026 (about 480 to about 4,400), flat since

Search-index estimate, re-pulled 16 November 2026. Single-vendor data in rounded bands, directionally solid. The figure read 4,400 in January 2026 and in every month since except May, at about 5,400, and April, which is missing. "Answer engine optimization" grew from a smaller base. It read about 2,400 from April to July and rose to about 3,600 by October.

Verified (working paper, added November 2026)

~40% - fewer organic clicks on desktop searches where an AI Overview appears, in a randomized experiment

Agarwal and Sen, 1,065 US participants on desktop Chrome, January to February 2026. A browser extension hid AI Overviews for a random group: 0.37 organic clicks per search with the overview against 0.62 without (38% in the April draft, 39.8% in the June draft, and a July revision we have not read). Not peer reviewed.

Verified (preprint, added November 2026)

-18.8 points - click-through to outside sites when every search is sent to AI Mode

Wang and colleagues, a preregistered experiment with 1,100 US Chrome users over one week in March 2026 (95% interval 15.3 to 22.2 points). AI Mode was 0.6% of these users' searches beforehand, so this measures full adoption, which has not happened.

Verified (added November 2026)

-23.1% - click-through on the most exposed sites, nine days after AI Overviews launched in France

Ahrefs compared 963 high-traffic French domains before and after the 22 July 2026 launch. Median click-through on the most exposed went from 2.29% to 1.76%; domains with almost no exposure rose 2.3%. A before-and-after comparison with a low-exposure comparison group, at a national launch. The 23.1% is the raw change for the most exposed sites. Adjusted for the comparison group's rise, Ahrefs puts it at 24.8%. Across all 963 domains the median fell 5.7%. It points the same way as the Agarwal and Sen experiment above. Nine days, one vendor, large sites only.

Verified: citation mechanics.

Verified

~12% - share of AI-cited pages that also rank in Google's top ten for the query

Ahrefs, July 2025 data. The average includes Perplexity at 28.6%; the other values sit between 6% and 9%. Ranking equity does not transfer automatically, and citation can be won without it. Measurements that also count an engine's sub-queries find more overlap, so read this as overlap with the original prompt.

Verified

Citation correlates with semantic relevance to the prompt, not brand authority directly

Measured via similarity between cited content and the query. An October 2026 University of Toronto preprint points the same way: whether ChatGPT's own search retrieved a page explained 32% of within-site citation variation, against 2.8% for page features. The most specific, most complete answer to the actual question beats the biggest brand more often than the older discipline would predict.

Verified

No citation lift from adding schema markup - matched before-and-after study of 1,885 pages

The correlation is real (roughly half of AI-cited pages carry schema, per Ahrefs, May 2026). Ahrefs' May 2026 study found no lift, with Google AI Overview citations declining slightly. Every page in it already had more than 100 AI Overview citations. An October 2026 preprint from the University of Toronto found no significant difference inside sites between pages with schema and their neighbors. Schema is a clarity layer, not a citation lever - the full argument is in How AI Verifies You're Real.

Verified

1.93% - share of the Reddit pages ChatGPT retrieved that it went on to cite

Ahrefs, April 2026, from 1.4 million ChatGPT prompts. Of everything ChatGPT retrieved but did not cite, 67.8% was Reddit pages. Ahrefs notes that some of those pages may never have been opened, so the figure does not show how much Reddit shaped the answers. Reddit's presence has also fallen: Ahrefs found it cited in about one in five US ChatGPT answers in March 2026 and about one in 43 by late September.

Verified

~26% - how much newer, by publication date, the pages AI assistants cite are than Google's organic results (AI Overviews contested)

Ahrefs, July 2025, about 17 million cited URLs: pages cited by ChatGPT, Perplexity, Copilot and Gemini were 25.7% newer than organic results by publication date, and 13.1% newer by last update. In the same data the top three citations in Google's AI Overviews were the oldest. Seer Interactive's June 2025 study of more than 5,000 URLs found the reverse. Both predate the model now behind AI Overviews. Refresh strategy has to name its target platform.

Verified - our data

89% / 1.93 of 5 / 0% - one profession's AI-search infrastructure, audited across 130 firms

Clarion's 2026 audit of independent RIA digital infrastructure: 89% of firms surface in AI search (through directories they don't control), mean schema coverage 1.93 of 5, and zero firms linking their own regulatory record. Full methodology in The AI Visibility Gap.

Verified (added November 2026)

50-59% - share of AI citations missing the next day. About three quarters return

Ahrefs asked five AI tools the same questions daily for 20 days and followed about 37,500 day-one citations. The 50-59% range covers Google AI Overviews, AI Mode and ChatGPT. Gemini and Copilot lost about 80%, a figure inflated by answers that carried no citations. A single capture of an AI answer is a sample of one.

Verified as dated (re-checked November 2026)

29.7% - Wikipedia's share of the 1,000 pages ChatGPT cited most (September 2025)

Ahrefs, from the same category table as the ~32% figure above. It is a share of those 1,000 pages, and Ahrefs' 2026 data shows Wikipedia appearing in far fewer ChatGPT answers since.

Verified as dated (re-checked November 2026)

28.6% - share of Perplexity's cited URLs in Google's top ten (July 2025)

Ahrefs. One of the five values averaged into the ~12% figure above.

Verified as dated (re-checked November 2026)

80% - share of assistant citations that do not rank in Google for the original prompt (July 2025)

Ahrefs, same study as the ~12%. The two count different things: pages in the top ten, and pages that do not rank for the prompt at all.

Verified as dated - a different measure

99.5% - AI Overviews whose sources include one or more of the organic top ten (August 2024)

seoClarity via Search Engine Land, 36,000 keywords. The denominator is not defined. Read as written, it counts overviews with one or more top-ten sources. The 38% below counts the share of all cited pages that rank in the top ten, so the two do not conflict. Two years old, and it predates the model now behind AI Overviews.

Verified as dated (re-checked November 2026)

"AI Overviews pull 76% of citations from the top 10"

This was Ahrefs' finding in July 2025, counted on the three most visible citations in each overview. Ahrefs measured again in March 2026, across about four million cited URLs, and got 38%. The two differ in method as well as date. The 76% was accurate for the most visible citations in July 2025 and does not describe AI Overviews now.

Refuted.

Each of these circulates widely and is contradicted by a source we can name. A claim is listed here only when that is true. A claim we cannot confirm goes on the hold list below.

Refuted (added November 2026)

"Schema markup increases AI citations by 30-40%"

Circulates with no method attached. The Ahrefs study above found no lift.

Circulating, unproven: the hold list.

These are not refuted. They are numbers we will not repeat as fact until their sourcing improves, and we suggest you hold them to the same standard.

Unproven - no measurement

"AI engines prioritize multi-publisher consensus over single-source claims"

The "consensus layer" theory, often used to sell placement-everywhere packages. We find no measurement of it, for or against. The measured correlate of citation is semantic relevance to the prompt.

Unproven - lab result

"GEO tactics boost AI visibility by up to 40%"

Re-checked November 2026. The sentence is in the abstract of the 2024 GEO paper. It is a benchmark result for sources already in the engine's context. A July 2026 survey of 45 GEO studies found no technique with a stable, cross-platform causal effect on being found. Accurate as a lab result. Unproven as a promise.

Unproven - one agency's client data

"Organic CTR has dropped 61%" (also circulated as 41%)

Re-checked November 2026. Both figures are Seer Interactive's, from 3,119 informational queries across 42 client organizations (published November 2025). Seer's April 2026 update showed click-through on AI Overview queries rising again. Ahrefs' comparison ends in December 2025, the month Seer's series bottomed, so it neither confirms nor contradicts the rebound.

Unproven - depends on the definition

"9 in 10 ChatGPT citations don't rank in the top 20"

Re-checked November 2026. Ahrefs measured in August 2025 that 9.85% of the URLs ChatGPT's searches returned were in Google's top 20 for the same search. AirOps, counting cited pages against the prompt and its sub-queries, found 55.8% in a top 20. The wording itself matches a July 2025 Semrush statement. Semrush did not say how it measured that figure.

Unproven - single vendor, contested

"Top-ranking pages lose ~58% of clicks to AI Overviews"

Ahrefs, February 2026, with a published method (300,000 keywords, December 2023 against December 2025). It is a correlation from one vendor. Seer Interactive's 2026 data shows click-through on AI Overview queries rising over the following months, which is a different measure. One randomized experiment, above, found about 40% fewer organic clicks on desktop searches where an AI Overview appears. That counts every organic result and the 58% counts the top position only, so it supports the direction and does not check this number.

Unproven - correlational

"AI-cited brands see 35% more clicks"

Seer Interactive, published November 2025: 0.70% organic click-through when cited against 0.52% when not. Seer states that it cannot show the citation causes the difference. Its April 2026 update, on a larger client set, puts the gap at 120% and repeats the caution.

Unproven - measures belief, not behavior

"91% of marketers name freshness as the top AI-citation factor"

An NP Digital survey of 100 content marketers, April 2026. It records what marketers believe drives citation. The verified freshness finding above is narrower and split by platform.

Unproven - category-wide

Every forward-looking prediction ("X% of queries will be answered by AI by year Y")

In our verification pass, no 12–24-month forecast in this field survived scrutiny of its basis. Analyst predictions are planning inputs, not measurements; treat every future-tense percentage accordingly.

How to read any AI-search statistic.

Four checks, applicable to every number you encounter in this field, including ours. First: is the primary source named, and is the citation to the source or to someone else citing it? Most circulation is citation-of-citation. Second: does the source measure behavior or belief? Surveys of marketers are belief. Third: is the claim causal or correlational, and does its seller know the difference? The schema case above is the canonical trap. Fourth: is it dated? The engines change quarterly; an undated statistic in this field is unfalsifiable. A number that passes all four is rare, which is why this page is shorter than the pages it competes with.

And a closing note on what the surviving numbers add up to, because a ledger without a reading is just a list. The demand-side figures say the answer layer is absorbing attention faster than most professionals have adjusted to. The mechanics say the new game scores specificity and verifiability over size, and scores them differently per platform. The winnable-third number says the opportunity is real and bounded. Put together: a narrow expert with verifiable claims and depth on the actual questions of their field is better positioned in this environment than they were in the ranking era, and a generalist buying volume is worse positioned. That is our reading of the evidence, not a measurement.

What to do with the numbers that survive is a separate question, and we've written it up in two places: the definitional ground in AEO vs GEO vs SEO, and the buildable mechanism in How AI Verifies You're Real. If you'd rather someone build it with you, that service exists, with its caveats attached in writing.

Questions people actually ask.

Is generative engine optimization a real thing?

Yes. The demand shift is measured, the term has an academic anchor, and attention to it grew roughly ninefold from late 2024 to early 2026 and has been flat since. What's frequently not real is the statistical case sold alongside it - hence this page.

What's the most reliable number here?

Pew's two measurements and the randomized experiment by Agarwal and Sen on the demand side. On the supply side, the ~12% overlap between cited pages and Google's top ten. The ~32% winnable share is useful and dated.

How much of AI search visibility is actually winnable?

About a third of ChatGPT's most-cited pages, in a September 2025 study. The source mix has shifted since, so treat it as a dated estimate. Within that share, the evidence favors specific, checkable expertise over brand size.

Does schema markup improve AI citations?

Correlated, not shown to cause citation. A matched study of 1,885 already-cited pages found no lift from adding it. It matters for a different reason - machine-checkable claims - covered in the walkthrough.

Why do refuted statistics keep circulating?

They sell. Dramatic numbers close deals, most citations are citations of citations, and re-checking is rare. Rare enough, in fact, that doing it is this studio's entire opening argument.


Mark Nelson

Founder, Clarion Studio

Mark builds machine-readable identity infrastructure for expert practices - entity graphs, structured data, and editorial systems, working at the intersection of strategy, design, and front-end engineering. Based in Lynden, Washington.

Published August 13, 2026 · Last re-checked November 2026

Read more about Mark·Start a conversation