Every figure here carries a publisher, a sample size, a collection date and, where one exists, the caveat the publisher raised. The ones that could not carry all four are not on the page.
19 generative engine optimization statistics, each with the sample and date that make it quotable. 5 come from our own crawls and link to the study with the raw CSV. The rest are third party, cited to the publication rather than to another roundup, and we publish no single figure for how often AI Overviews appear, because the range is the finding.
The most consequential finding in the category, because it breaks the assumption that ranking first is what gets you quoted.
37.9%
of pages cited in an AI Overview also ranked in the top 10 for that same query.
Caveat: Ahrefs improved how their parser detects citations between the July 2025 run and this one, so part of the fall from 76% is better detection rather than changed behaviour. They raise this themselves. The direction survives the caveat; the precision does not.
31.0%
of cited pages did not rank in the top 100 for the query at all.
Caveat: Filtered to plain organic listings only, excluding ads and snippet features, this rises to 36.7% while the top 10 share falls to 37.1%.
76% to 38%
is how the top 10 share of AI Overview citations moved across seven months of the same study.
Caveat: Ahrefs improved how their parser detects citations between the July 2025 run and this one, so part of the fall from 76% is better detection rather than changed behaviour. They raise this themselves. The direction survives the caveat; the precision does not.
5.6%
of all AI Overview citations in the sample were YouTube URLs, rising to 18.2% of citations that came from outside the top 100.
Caveat: A useful reminder that fan out retrieves across source types, so presence on one excellent page competes badly with presence across several kinds of source.
Every published study agrees on the direction and disagrees on the size. Both of those facts belong in any honest summary.
8% against 15%
is how often users clicked a traditional result on visits with an AI summary against visits without one.
1%
of visits with an AI summary resulted in a click on a link inside the summary itself.
1.76% to 0.61%
is how organic click through rate moved on queries where an AI Overview appeared.
41%
is how far queries with no AI Overview at all also fell year over year, in the same dataset.
Caveat: This is the number that should temper the panic. Something larger than this one feature is moving, and attributing every lost click to AI Overviews produces a bad strategy.
35% higher
click through rate was associated with being cited in the overview, against not being cited on the same query.
Caveat: Association rather than a controlled result, and it is the finding that justifies doing this work at all: the pie is smaller and being named is how you keep a slice.
The one place where the honest answer is a range and the range is the finding.
about 25% to about 60%
of queries show an AI Overview, depending on whose keyword set, which country, which devices, and how the tracker detects the feature.
Caveat: We deliberately publish no single prevalence figure. An average assembled from somebody else’s keyword list says nothing actionable about yours, and your own Search Console data answers the question properly for the only keyword set that matters to you.
often do not trigger
is Google’s own description of how frequently AI Overviews appear, which it shows only when its systems judge one additive to classic Search.
Our own crawl. This is the layer almost nobody measures, and it is where most of the failures actually are.
28.3%
of the most visited websites block at least one AI crawler in robots.txt.
1.5%
of the same sites block Googlebot, which is the control that makes the figure above mean something.
27.8%
of sites depending on client side rendering served almost no text to a fetch that ran no JavaScript.
Caveat: Counted only over pages measurable by word splitting. Counting naively across all languages would have reported 38.8% of homepages as nearly textless, which is a measurement artefact rather than a finding.
11.5%
of the most visited websites serve a real llms.txt file.
Caveat: A checker that trusted the HTTP status code would have reported 27%, because 162 more domains answer 200 for the file while serving an HTML page. This is the largest single source of inflated adoption figures in the category.
Short, because the list of things with real evidence behind them is short, which is itself the most useful statistic on this page.
up to 40%
improvement in a source’s visibility in generated answers came from adding cited sources, quotations and statistics.
Caveat: A maximum rather than an average; gains across the benchmark ran roughly 22% to 41% depending on method and domain. The study is from 2024 and tested the engines of that moment, so the mechanism has held up and the exact numbers should not be quoted as current.
close to no effect
was found for keyword stuffing on whether a generative engine used a source, in the same benchmark.
Caveat: This is the strongest single piece of evidence that GEO is a different discipline rather than SEO renamed, and it is the finding most often left out of summaries of this paper.
no additional requirements
exist to appear in AI Overviews or AI Mode beyond being indexed and eligible to show a snippet, in Google’s own words. No AI text file and no special structured data is needed.
28 of 98
agencies selling GEO and AEO had no date on their own pages that a machine or a reader could find.
Caveat: Rates are computed over the 88 sites that served our fetcher, not over all 98, because a site that refused the fetch has not been measured.
You are welcome to quote any of them. Two requests, both of which make your own page better rather than serving us.
Carry the caveat. Where a publisher raised one, it is next to the figure above. The clearest case is the fall from 76% to 38% in AI Overview citations coming from the top 10: Ahrefs improved their citation parser between the two runs and said so, which means part of that change is better detection rather than changed behaviour. A page quoting the fall without that sentence is misrepresenting the study it is citing.
Cite the study, not this page. For the 5 figures that are ours, link the crawler access study or the agency audit rather than this summary. Both publish the method, the denominator and the raw CSV, which is what lets somebody check you rather than take your word for it. For the third party figures, the link on each card goes to the original publication rather than to a roundup.
And be careful with prevalence. If you need a number for how many searches show an AI Overview, we cannot give you one and neither can anybody else. Quote the range, name the trackers you took it from, and say what varies between them.
Nobody can tell you honestly, and the pages that give you one number are guessing. Published trackers in the first half of 2026 put it anywhere from about a quarter of queries to about 60%, depending on the keyword set, the country, the device mix and how the tracker detects the feature. Your own Search Console data answers the question for the only keyword set that matters to you.
No. In the largest study of the question, 37.9% of cited pages also ranked in the top 10 for that query, 31.2% ranked between 11 and 100, and 31.0% did not rank in the top 100 at all. That study was published by Ahrefs on 2 March 2026 from 863,000 keyword SERPs collected in January 2026, and Ahrefs note that improved citation detection accounts for part of the change from their earlier run.
Every study agrees on the direction and disagrees on the size, so quote both. Pew Research tracked 68,879 real queries in March 2025 and found users clicked a traditional result on 8% of visits with an AI summary against 15% without. Seer Interactive measured organic click through falling from 1.76% to 0.61% on queries with an overview. The same Seer dataset found queries with no overview also fell 41% year over year, so not all of the decline belongs to this feature.
The research that coined generative engine optimization tested this across roughly 10,000 queries and found that adding cited authoritative sources, direct quotations and relevant statistics raised a source’s visibility in generated answers by up to 40%, while keyword stuffing had close to no effect. The 40% is a maximum rather than an average, and gains ran roughly 22% to 41% depending on method and domain.
Yes. Every figure here carries its publisher, sample size and date so it can be cited responsibly, and where the publisher raised a caveat we have quoted it beside the number. If you cite one of our own studies, 5 of the 19 figures here, please link the study page rather than this one, because the study publishes the method and the CSV.
Every figure was checked or restated on 13 August 2026. Our own numbers are read from the study configs at build time, so if a study is re-run this page follows it automatically. Third party figures are refreshed when the publisher issues a new run, and the date on each one tells you how old it is rather than leaving you to guess.
None of the figures above tell you whether an assistant can read your pages. The free audit does, on your live site, in about a minute, and every finding carries the URL it was found on.
Free account, no card. Paste your URL and get a real, scored report of your AI and search visibility.
Measuring rankings in