BlogFindings
How often do AI Overviews appear? One headline said over 80 percent, and the same method gave 40.4 percent in Germany
Search Engine Roundtable reported on 26 October 2026 that AI Overviews now show up for over 80 percent of Google searches, attributing the figure to recent data from Semrush, SE Ranking and seoClarity. Lantad opened all three sources on 27 October 2026. Semrush published 88 percent of 400 United States desktop commercial keywords, SE Ranking published 82.5 percent of 1,000 United States brand keywords and 40.4 percent for the same method on 1,000 German ones, and the seoClarity figure of over 62 percent is an interview remark with no study behind it. Google publishes no figure at all.
Lantad has measured none of this. The scanner reads what a crawler can fetch from a site and grades what is readable, which is the site side of the problem, and nobody outside Google can count how many searches trigger an AI answer. So this post reports other people's studies, names what each one sampled, and then turns to the part a reader can actually check on their own server. That second half is where AI visibility work either happens or does not, and it is indifferent to whether the true rate is 40 percent or 97.
In short
- How often do AI Overviews appear is not the question any of the three studies behind the over 80 percent headline set out to answer: Semrush measured 400 United States desktop commercial keywords, SE Ranking measured 1,000 United States brand keywords, and the seoClarity figure was an interview remark, so the three populations are commercial intent, branded intent and an unstated query set.
- SE Ranking checked 1,000 United States and 1,000 German brand keywords three times each on desktop in October 2026 and published AI Overview presence of 82.5 percent in the United States against 40.4 percent in Germany, so one method run on one sample in one month halved its own headline figure when the country changed.
- Semrush tracked 400 commercial keywords across banking, software and ecommerce from September 2025 to August 2026 and published coverage rising from 45.75 percent to 88 percent, with a volume weighted figure of 97.08 percent for August 2026 because the 12 percent of keywords carrying no AI Overview accounted for under 3 percent of the search volume.
- Google's own AI features documentation carried a last updated date of 2025-12-10 UTC when Lantad read it on 27 October 2026, publishes no rate for how often AI Overviews appear, and states only that they often do not trigger, so the one party that holds the exact number releases none of it.
- Lantad measured 132 of 1,083 readable corpus home pages sending a snippet directive on 7 October 2026 and exactly 3 of those set a value that limited anything, which is the documented lever over what an AI Overview may quote from a page whatever the appearance rate turns out to be.
How often do AI Overviews appear, and what did each study actually count?
The headline is a post on Search Engine Roundtable dated 26 October 2026, titled Over 80% Of Google Searches Have AI Overviews & Growing, at www.seroundtable.com/over-80-of-google-searches-have-ai-overviews-42372.html. Read on 27 October 2026 it says AI Overviews now show up for over 80 percent of Google Searches, and attributes that to recent data from Semrush, SE Ranking and seoClarity. The aggregate figure itself is not assigned to any one of the three, and no sample or population is stated for it anywhere in the post.
That matters because the three underlying figures are not measurements of the same thing. Semrush measured commercial keywords, which are the queries where somebody is close to buying. SE Ranking measured branded keywords, which are queries containing a company name. The seoClarity figure, as the post presents it, covers queries with no geography or sample size given. Those are three distinct slices of search demand, and all three are slices chosen for commercial relevance rather than drawn at random from everything typed into Google. Averaging them, or rounding up past the highest of them, produces a sentence about Google searches that none of them supports.
This is a recurring shape rather than one bad week. Google announces a change to its AI answers, the trade press reaches for a number, and the number turns out to be measured on a panel. This blog made the same point about Google's own announcement of 15 October, where five named link changes arrived with no quantity attached to any of them and the only figure in the post was a multiple with no base. The difference here is that the figures do exist and are carefully documented by the people who produced them. It is the summary that drops the sample, and the sample is the whole of the information.
None of this makes the studies wrong or the trend imaginary. Coverage of commercial queries clearly rose a lot over the last year, and both published studies say so with their working shown. The useful question for anyone doing answer engine optimization is narrower and more answerable: on the queries that matter to me, in the country I sell in, how often does an AI answer sit above the results, and what would I change if I knew. The three figures below are the best available evidence on the first part, and they disagree by more than two to one.
| Source | Population measured | Sample size | Country | Device | Period | Method published |
|---|---|---|---|---|---|---|
| Semrush, 21 Oct 2026 | Commercial keywords in three verticals | 400 keywords | United States | Desktop | Sep 2025 to Aug 2026 | Yes |
| SE Ranking, 22 Oct 2026 | Brand keywords across 17 industries | 1,000 per country | United States and Germany | Desktop | October 2026 | Yes |
| seoClarity, via interview | Queries, not further defined | Not stated | Not stated | Not stated | About a month before 26 Oct | No |
| The headline figure | Stated as Google searches | None given | None given | None given | None given | No |
| All Google searches | Holds the number | Global | All | Continuous | Publishes none |
Semrush tracked 400 commercial keywords, not Google search
The Semrush study is published at www.semrush.com/blog/ai-overviews-commercial-modifiers-study/, dated 21 October 2026, bylined Luke Harsel with Nikola Manojlovic as contributor. Semrush is a competitor of ours, since the same company ships an AI visibility product, so this post names it in prose and does not link it, and the comparison page we maintain is at our write-up of the Semrush AI visibility product. The study itself is careful work and says plainly what it did.
It tracked 400 commercial keywords across banking, software and ecommerce topics, monthly, from September 2025 through August 2026, on United States desktop data. The keywords are grouped into five commercial modifiers: alternatives, brand versus brand, best X for Y, pricing and cost, and reviews. Keyword weighted coverage rose from 45.75 percent in September 2025 to 88 percent in August 2026. The modifier spread in August 2026 ran from 82.05 percent to 96.12 percent, with best X for Y highest, and the largest single rise was reviews, up about 59 points from roughly 30 percent to nearly 90. By industry, September 2025 ran from ecommerce at 32.62 percent to software at 68 percent, and by August 2026 all three verticals sat between 85 and 91 percent, with ecommerce at 85.11.
One figure in that study deserves separating from the rest, because it is the one most likely to be quoted without its definition. Semrush also reports a volume weighted coverage of 97.08 percent for August 2026, and explains it: the 12 percent of keywords with no AI Overview accounted for under 3 percent of the search volume in the set. That is a real and interesting result about where the volume sits inside a commercial panel. It is not 97 percent of Google searches, and read as such it would be wrong by a wide margin.
So the honest reading of the Semrush number is specific and still useful. On 400 commercial queries in three industries, in the United States, on desktop, an AI Overview was present on 88 percent of them by August 2026, up from under half a year earlier. Anyone selling banking, software or ecommerce products to a United States audience should treat the AI answer as the default state of their commercial result pages. That is a strong conclusion about generative engine optimization for those buyers, and it says nothing about navigational queries, informational queries, mobile, or any country other than one.
The same method gave 82.5 percent in the United States and 40.4 percent in Germany
The second study is the one that settles the argument, because it ran a single method over two countries in the same month. SE Ranking published it at visible.seranking.com/blog/ai-overviews-branded-searches/ on 22 October 2026. It took 1,000 United States and 1,000 German brand keywords across 17 industries, split in each country into 333 small business, 333 mid market and 334 enterprise keywords, and checked every keyword three times on desktop during October 2026.
The United States figure is the 82.5 percent that fed the headline, averaged across the three checks. The German figure, from the same design, the same month and the same industries, is 40.4 percent. Consistency moved the same way: 75.6 percent of United States keywords showed an AI Overview on all three checks against 29.6 percent in Germany, and 22.2 percent of German keywords flipped between showing one and not showing one against 11.4 percent in the United States. Some industry gaps are extreme. Industrial and business to business ran 84.7 percent in the United States against 10.2 percent in Germany, and energy and utilities 83.7 against 9.3.
Two other cuts inside the United States sample are worth keeping, because they contradict the instinct that bigger queries get more AI answers. Presence fell as volume rose: 85.2 percent on keywords under 50,000 monthly searches against 63.3 percent on those over a million, a gap of 21.9 points. By company size, mid market brands drew AI Overviews on 78.7 percent of their keywords and enterprise brands on 70.7 percent. A rate measured on a panel of large well known brands is therefore not an upper bound on what a smaller site sees. It can be the floor.
The study also reports something directly useful rather than merely cautionary, and it lines up with work this blog has already covered. Of 1,995 brand site citations in the United States AI Overviews it analysed, 62.2 percent were not in the organic top 10 for the same query, rising to 92.6 percent for product, service and category pages against 16.5 percent for home pages. That is the same phenomenon a Washington University in St. Louis team found over 55,393 trending queries, where 29.8 percent of cited domains appeared nowhere on the first page of results for the query. Two independent samples, two different figures, one consistent direction: the pool an AI answer draws from is not the top ten. SE Ranking also found that 40.5 percent of United States citations came from brand managed sources, which is the share a site owner can influence, and this blog measured on 26 September 2026 that 412 of 1,080 readable corpus home pages named their own site in neither of the two places Google reads first.
The third figure behind the headline has no published study
The third source named in the headline post is seoClarity, and the figure attributed to it is over 62 percent of queries, with a range of 55 to 84 percent depending on industry. Read on 27 October 2026, the post presents this as something Mark Traphagen of seoClarity told the author about a month earlier. It links the company home page and no study. Searching for a seoClarity publication carrying that figure on 27 October 2026 returned none.
That is worth stating precisely, because it is not an accusation of anything. A vendor telling a journalist what its data shows is ordinary, and Traphagen may well be describing a large and careful measurement. The problem is downstream: the figure cannot be audited, its sample is unknown, its date is approximate, and it is the lowest of the three numbers feeding a headline that rounds upward past all of them. A reader who wanted to check the weakest link in the chain has nothing to open.
Lantad is a vendor too and the same standard applies here, which is why this blog publishes its method with every figure and says where its own numbers come from. It has reported its own tooling disagreeing with itself: on 26 October 2026 the same 31 stored answers scored 33.3 percent share of voice on one shipped surface and 36.9 on the other because the two call paths used different denominators. A vendor figure with no method attached could hide exactly that and nobody would know.
There is also a plainer reason to be careful with category figures from tools in this market, which is that the tools themselves move faster than anybody's screenshots. Reading the pricing pages of seven GEO tools on 26 October 2026, this blog found five of seven published a price against three of six a month earlier, with two changing verdict inside 32 days. If a published price can appear, disappear and reappear across three readings, an unpublished prevalence figure quoted from memory a month after a conversation deserves the same scepticism, and the people best placed to supply the missing method are the ones who produced it.
-
Semrush, 88 percentCheckable Study published with sample size, verticals, country, device and a 12 month window stated. -
SE Ranking, 82.5 percentCheckable Study published with 1,000 keywords per country, 17 industries, three desktop checks and a limitations section. -
SE Ranking, 40.4 percentCheckable Same publication. The German half of the same design, which the headline did not carry. -
seoClarity, over 62 percentNot published Presented as an interview remark about a month old. No study, sample size, country or device stated. -
Over 80 percent of Google searchesUnsupported as written Higher than two of the three inputs and describes a population none of them sampled. -
Google's own rateWithheld Google holds the exact figure and publishes no number for it.
Google publishes no figure, and says AI Overviews often do not trigger
The one organisation that can count this exactly is Google, and it publishes nothing. Read on 27 October 2026, Google's AI features and your website documentation carried a last updated date of 2025-12-10 UTC and gives no rate, no share of queries and no trend. What it does say is that AI Overviews are shown only when Google's systems determine they add value beyond classic Search, and that they often do not trigger. That sentence is the closest thing to an official statement on frequency, and it is a direction rather than a quantity.
The same page is more specific about the things a site owner controls, and on those it is unambiguous. A page is eligible to appear as a supporting link if it is indexed and eligible to be shown in Google Search with a snippet, fulfilling the Search technical requirements, and it states there are no additional technical requirements. There is no AI Overviews markup, no file to publish and no setting on your site that increases the rate at which an AI answer appears for a query. The only site level controls it names are robots.txt for Googlebot crawling, noindex, and the snippet directives.
Those snippet directives are where the documentation gets precise, and they are the part worth reading closely. Google's robots meta tag documentation, last updated 2026-03-24 UTC when read on 27 October 2026, states that nosnippet applies to all forms of search results including AI Overviews and AI Mode and will also prevent the content being used as a direct input for AI Overviews and AI Mode, and that max-snippet will also limit how much of the content may be used as a direct input for those surfaces, except where a publisher has separately granted permission for use of content.
One control that readers often look for is not on the site at all. Excluding a site from AI Overviews and AI Mode entirely is an account setting in Search Console rather than a file or a tag, which this blog wrote up when the setting appeared: the AI Overviews opt out does not live on your site. That has a consequence for every tool in this category, Lantad included. No external scan of a domain can tell you whether that switch is set, so no scanner can give a complete account of a site's AI Overviews exposure, and any that implies otherwise is overselling.
- A published rate for how often AI Overviews appear No figure, share or trend anywhere on the AI features page. It states only that AI Overviews often do not trigger.
- A stated eligibility condition The page must be indexed and eligible to be shown in Google Search with a snippet, fulfilling the Search technical requirements.
- Additional technical requirements beyond that The documentation states there are none, and names no AI Overviews specific markup or file.
- nosnippet as a control over direct input Documented as preventing content being used as a direct input for AI Overviews and AI Mode, and as applying to all forms of search results.
- max-snippet as a cap on direct input Documented as limiting how much content may be used as a direct input, except where a publisher has separately granted permission.
- A site side switch to exclude a site entirely The exclusion is a Search Console account setting, so it is not a file or tag and cannot be read by any external scan.
What a site owner can actually check, whatever the rate is
The prevalence figure is interesting and not actionable. The eligibility condition is boring and entirely actionable, and it is the same condition whether AI Overviews appear on 40 percent of your queries or 97. A page has to be indexed and snippet eligible, which means a crawler has to reach it, receive a readable document, and find no directive telling it not to quote what it read. Each of those three is a fact about your own server on a given date, and this blog has counted all three across a fixed corpus of 1,419 hostnames.
Start with the directive, because it is the only documented lever and almost nobody has pulled it. On 7 October 2026 Lantad read the home page of all 1,419 corpus hostnames with no JavaScript executed and found 132 of the 1,083 that answered with HTML sent a snippet directive, of which 129 set max-snippet to minus one, which Google defines as no limit. Exactly 3 sites in the corpus sent a value that limited anything. The directive is present on about one readable page in eight and active on three pages in 1,083, so the overwhelming majority of sites have expressed no position at all on how much of their content an AI answer may use.
The related attribute goes wrong in a quieter way. On 2 October 2026 a run over the same corpus found 76 of 225 data-nosnippet attributes sat on an element Google does not support, which means the author believed they had excluded a block and had not. A control that silently does nothing is worse than no control, because it stops the owner looking. In the other direction, a capture of 120 hostnames on 6 September 2026 found that of the 81 answering HTTP 200, none carried a nosnippet directive blocking their own eligibility, so accidental self exclusion by that route was rare in that sample.
Then there is the plainest failure of all, which is the page not arriving. Repeated runs over this corpus put that loss between a quarter and a third of sites on a single stateless request. On 26 October 2026, 439 of the 1,419 failed at least one of four gates drawn from OpenAI's own crawler documentation, through a robots.txt refusal, a non 200 home page, HTML with no extractable words or a missing title. On 13 October 2026 a full scan with a real browser attached graded 972 of the 1,419, leaving 447 ungraded, of which 287 refused a plain fetch, 35 blocked the browser, 35 timed out and 21 failed to resolve in DNS. The population and the method behind both runs are set out in the crawlability study. Where the words arrive only after JavaScript runs, the gap is the prose parity problem, which took 60.7 percent of every point lost across those 972 graded sites. A fast way to see what a stateless fetch of one page actually returns is the what GPTBot sees view, and the full account of how the grading works is on the methodology page.
Flow: Crawler reaches the page (measurable) to Readable words returned; Readable words returned (measurable) to Snippet directive permits quoting; Snippet directive permits quoting (stated condition) to Indexed and snippet eligible; Indexed and snippet eligible (inside Google) to Query triggers an AI Overview; Query triggers an AI Overview (inside Google) to Page chosen as a supporting link.
What this post does not say
It does not say how often AI Overviews appear. No figure here is Lantad's, no engine was queried for this post, no search result page was captured and no query panel was built. Everything in the first four sections is other organisations' published work, read at source on 27 October 2026 and reported with the sample attached. If the three studies are each accurate about their own panels, the true rate across all Google searches is still unknown, and it is probably lower than any of them, because commercial and branded queries are exactly the slices where an AI answer is most likely and the long tail of navigational and trivial queries is absent from all three.
It also does not say the trend is overstated. Both published studies show large increases, Semrush over a 12 month window on a fixed panel and SE Ranking through consistency across repeat checks, and this blog has reported related external findings that point the same way: Google said on 15 October 2026 that it is placing more links inside its AI answers rather than only beneath them. The criticism here is of one unsourced aggregate sentence, not of the measurement work behind it.
Two things the two vendors did that Lantad does not do are worth naming plainly. Both measured the result page side, which means tracking whether an AI answer appears for a given query over time, across countries, industries and volume bands. Lantad measures none of that and has no such instrument: it reads sites, not search engine result pages. For a buyer whose question is how often an AI Overview appears on my queries in my market, SE Ranking's two country design or Semrush's 12 month panel is the right kind of evidence and this product supplies nothing comparable. SE Ranking also published a limitations section with its study, naming the desktop only snapshot and the small industry subgroups, which is the practice this blog is arguing for.
Finally, what is measured on the site side carries its own limits, and they are the standing ones for this corpus. The 1,419 hostnames are two committed sampling frames stratified by platform and sector, not a random draw from the web. The unit is usually a single page, the apex home page, fetched once from one network location, which cannot separate a standing policy from a transient error. Every request is sent as LantadBot rather than as Googlebot, and since edge rules commonly verify crawlers by address, a site that refuses this scanner may admit Google, which none of these runs tested. The crawler that does the fetching documents itself on the bot page, the vocabulary is on the AI crawler entry, and what engines actually say about a brand is a separate measurement on what AI says.
-
AI Overview appearance rateReported, not measured Every rate here is Semrush's, SE Ranking's or an interview remark, each with its sample named. -
Google's official rateDoes not exist publicly The AI features documentation gives no figure and says they often do not trigger. -
Snippet directive adoptionMeasured by Lantad 132 of 1,083 readable corpus home pages on 7 October 2026, with 3 setting a limiting value. -
Inert data-nosnippet attributesMeasured by Lantad 76 of 225 sat on an unsupported element on 2 October 2026. -
Whether a site is excluded in Search ConsoleNot measurable An account setting, invisible to any external scan of the domain including this one. -
Result page tracking by query and countryNot a Lantad capability Both vendors named here do this and this product does not.
Lantad
Published .
A number went round the search industry this week: AI Overviews now appear on over 80 percent of Google searches. It is the kind of figure that gets repeated in pitch decks for a year, and it is worth asking how often do AI Overviews appear with the question put properly, because the answer depends entirely on which searches you count. The three sources behind that headline measured three different populations, none of which is Google searches, and one of them is not published anywhere. Reading them in order is a short lesson in why a prevalence figure for AI Overviews needs its sample attached or it means nothing.
Common questions
How often do AI Overviews appear on Google searches?
Nobody outside Google publishes a defensible figure for all searches, and Google publishes none. The best available evidence covers slices: Semrush found AI Overviews on 88 percent of 400 United States desktop commercial keywords in August 2026, and SE Ranking found them on 82.5 percent of 1,000 United States brand keywords in October 2026 and 40.4 percent of 1,000 German ones using the same method. Commercial and branded queries are the slices where an AI answer is most likely, so the rate across all searches is probably lower than any of those figures.
Where does the over 80 percent of Google searches figure come from?
A Search Engine Roundtable post of 26 October 2026, which attributes it to recent data from Semrush, SE Ranking and seoClarity without assigning it to any one of them and without stating a sample or population. It is higher than two of its three inputs, and the third input, seoClarity's over 62 percent, is presented as an interview remark with no published study behind it.
Why did the same study find 82.5 percent in the United States and 40.4 percent in Germany?
SE Ranking's study of 22 October 2026 ran one design over two countries in October 2026, 1,000 brand keywords each across 17 industries, checked three times on desktop. The United States result was 82.5 percent and Germany 40.4 percent. The study does not attribute the gap to a cause, but it reports the same split in consistency and in individual industries, with industrial and business to business at 84.7 percent in the United States against 10.2 percent in Germany.
Can I change how often an AI Overview appears for my queries?
Not according to Google's own documentation. Its AI features page, last updated 2025-12-10 UTC, states that a page is eligible if it is indexed and eligible to be shown in Search with a snippet and that there are no additional technical requirements, and it names no markup or file that affects whether an AI Overview triggers. What a site controls is whether a crawler can reach and read the page, and the nosnippet and max-snippet directives that govern how much of it may be used as direct input.
See what AI can read on your site
Run a free scan and get a graded report of exactly what AI crawlers can and cannot read, with ranked fixes.
AI visibility
AI visibility is whether AI systems can access, read, and faithfully represent your web pages, the prerequisite for being cited in an AI answer.
Generative Engine Optimization (GEO)
Generative Engine Optimization (GEO) is the practice of structuring and publishing content so that generative AI engines can fetch it, read it in full, and cite it when they answer a question.
Answer Engine Optimization (AEO)
Answer Engine Optimization (AEO) is the practice of making a page readable, allowed, and structured enough that AI answer engines can use it as a source when they compose an answer.
Nearly a third of AI Overview citations are not on page one
A Washington University in St. Louis team issued 55,393 trending queries over 40 days and found that 29.8 percent of the domains an AI Overview cites appear nowhere on the first page of results for the same query.
Google AI Mode: 132 of 1,083 home pages sent the directive that governs it, and 3 limited anything
Google's robots meta tag documentation states that nosnippet prevents content being used as a direct input for AI Overviews and AI Mode, and that max-snippet limits how much may be used. Lantad read the home page of all 1,419 hostnames in this repository's two committed corpus seed files on 7 October 2026 with no JavaScript executed. 1,083 answered HTTP 200 with HTML. 132 sent one of those two directives. 129 of them set max-snippet to -1, which Google defines as no limit at all.
AI Mode citations: five link changes on 15 October, and the only number in the post is a multiple with no base
Google published a Search post on 15 October 2026 naming five changes to how AI Mode and AI Overviews link out to the web, three of them described as tests rather than rollouts. The post gives no number or percentage for the increase in inline links it announces. The only figure in it is that Preferred Sources users have grown more than 7x since the start of the year, stated with no base count and no method. Lantad has measured none of these surfaces and does not claim to.
