Ask how often AI Overviews appear and you get about seven different answers, running from roughly 13 percent of searches to roughly 48 percent. Every one of them is real, and not one of them describes your site.
I captured Google’s own answer to this question in two markets on the same day, and it did not agree with itself. So here is what those published figures actually count, and how to get a coverage number for your own queries instead of borrowing somebody else’s.
Key Takeaways
- Published estimates put AI Overviews on roughly 13% to 48% of Google searches, and most of that gap is about what each study counted, not about Google changing.
- Every figure is one company measuring its own keyword panel, on its own dates, in its own markets, so it describes that panel rather than your pages.
- Google’s own AI Overview answered this question differently in two markets on the same day: 13% to 48% of all search queries in the United States, 21% to 48% of general search queries in the United Kingdom, both captured signed out on 30 July 2026.
- Query type moves the number far more than any headline average. Ahrefs found informational queries account for 99.9% of AI Overview appearances, while shopping queries trigger one only 3.2% of the time.
- In our own ecommerce set the AI Overview fired on 420 of 420 usable cells across 21 queries and four markets, which is nowhere near 3.2%.
- On one query in the United Kingdom it has not fired for 13 consecutive rounds, while firing in the US, Canada and Australia on every one of those nights.
- The only coverage rate that can change a decision is the one you measure on your own query list, and that takes an afternoon.
How often do AI Overviews appear in Google search?
Published estimates put AI Overviews on roughly 13% to 48% of Google searches, depending on the tracking tool, the keyword set and the date. The newest readings sit near the top of that band: Similarweb reports about 43% of all queries and Semrush about 48%, both as reported by Search Engine Roundtable on 30 July 2026, while Ahrefs’ November 2025 analysis of 146 million search results put it at 20.5% of all keywords.
That is the honest answer, and it is a range rather than a number for a reason. Here is every reading I could verify against its own live source this week, sorted by size rather than by date, with what each one actually counted sitting next to it.
| Reading | What it counted | Measured | Market | Who measured it |
|---|---|---|---|---|
| 100% | 420 usable cells, 21 ecommerce queries | Jul 2026 | US, CA, UK, AU | WpConsults, our own captures |
| ~48% | all queries | Jul 2026 | not stated | Semrush Sensor, via Search Engine Roundtable |
| ~48% | tracked queries | Feb 2026 | not stated | BrightEdge, via SQ Magazine |
| ~43% | all queries | Jul 2026 | not stated | Similarweb, via Search Engine Roundtable |
| 20.5% | all keywords, 146 million search results | Nov 2025 | not stated | Ahrefs, via Search Engine Journal |
| 15.69% | more than 10 million terms | Nov 2025 | not stated | Semrush, via SQ Magazine |
| 13.14% | all U.S. desktop searches, 10 million+ keywords | Mar 2025 | US desktop | Semrush and Datos, via Search Engine Land |
| 0% | one query, 13 consecutive rounds | Jul 2026 | UK only | WpConsults, our own captures |
Notice the shape of that spread. The two figures at the extremes are both ours, both first-hand, and both measured in the same month as the 43 and 48 percent headlines.
Why do AI Overview statistics disagree so much?
They disagree because each study counted something different, and almost none of them says so in the headline. A coverage percentage is a property of the measurement, not a property of Google, so two honest teams can publish figures three times apart and both be right about their own data.
Four things move the number, and you can usually spot which one is doing the work:
- The denominator. “All U.S. desktop searches” and “tracked queries” are not the same population. The most carefully scoped figure in the whole set is Search Engine Land’s 13.14% of all U.S. desktop searches for March 2025, drawn from a Semrush and Datos analysis of more than 10 million keywords, and it is scoped that tightly because someone did the work of saying what was counted. Worth knowing while you read it: Search Engine Land is owned by Semrush and says so at the foot of the article, which is disclosure done properly rather than a reason to discount the number.
- The date. Coverage has climbed fast. Reporting the 43 percent figure at Search Engine Roundtable, Barry Schwartz
Barry SchwartzFounder, Search Engine RoundtableFounder of Search Engine Roundtable and a long-running search industry news reporter, also News Editor at Search Engine Land.XLinkedInSearch Engine Roundtable notes that he “covered these percentages a year ago and it was at about 13% to 20%.” One publication, one measurement habit, a year apart, and the number roughly doubled. - The market. Almost every figure you will read is either US-only or silent about geography, and silence usually means a US-weighted panel. That matters more than it sounds, and the next section is why.
- The mix of query types. This is the biggest one and it is the reason a blended average describes almost nobody. If your keywords are questions, your real exposure sits far above the headline. If they are product searches, it may sit far below it.
You can watch this happen inside a single page. The SQ Magazine statistics roundup that reports about 48% of tracked queries from BrightEdge for February 2026 also carries Semrush’s reading of about 15.69% by November 2025, without reconciling the two. Neither figure is wrong. They are answers to two different questions wearing the same sentence.
Google’s own AI Overview gave two different answers in two markets
On 30 July 2026 I searched the same question twice, signed out, once with the market set to the United States and once to the United Kingdom, and read both screenshots by eye. Google’s AI Overview answered with 13% to 48% of all search queries in the US and 21% to 48% of general search queries in the UK.
Three things changed between those two frames. The floor of the range moved from 13 to 21. The wording of the denominator moved from “all search queries” to “general search queries”. And the panel above the answer said 9 sites in the US and 5 in the UK.
Now the honest limits, because this is the part most write-ups would skip. Google’s AI Overview is not reporting Google’s telemetry here. It is summarizing the same vendor blog posts you and I can read, so what these two frames prove is that the summary of the evidence is market-dependent, not that UK coverage is lower than US coverage.
The site counts are worth even less as data. Google’s “N sites” chip is a label on a panel, not a verified count of the sources behind the answer, and I am reading it off a single frame rather than a complete enumeration. Treat it as a difference worth noticing and nothing more.
Even so, if you have been building a plan around one borrowed percentage, this is the cheapest possible demonstration of the problem. The most authoritative-looking source on the query contradicted itself across a border on the same afternoon.
What triggers a Google AI Overview?
Query intent and query length trigger AI Overviews more than any other factor. In Ahrefs’ analysis of 146 million search results, reported by Search Engine Journal in November 2025, informational queries accounted for 99.9% of AI Overview appearances, question-based queries triggered one 57.9% of the time against 15.5% for non-questions, and queries of seven or more words triggered one 46.4% of the time against 9.5% for single-word searches.
The category splits in that dataset are just as wide, and they are where a blended average stops being useful:
- Science 43.6%, health 43.0%, medical YMYL 44.1%. Explainer-heavy topics sit at roughly double the overall rate, which is why publishers of educational content feel this hardest.
- Shopping 3.2%, the lowest in the dataset, and real estate 5.8%. On Ahrefs’ panel, commercial product searches almost never triggered one. Hold on to that number, because our own data disagrees with it sharply and the next section is that disagreement.
- Local queries 7.9%, against 22.8% for non-local. If your traffic is mostly “near me” searches, the headline figures overstate your exposure by roughly a factor of three. We cover that lane separately in our piece on local SEO and AI Overviews.
- Very newsy keywords 6.3%, against 20.7% for everything else. Google appears to hold back on time-sensitive topics, which fits a system that has to be right about facts that change hourly.
Read those splits against your own keyword mix before you read any headline percentage. A site whose queries are mostly seven-word questions about health is living in a completely different search result from a store whose queries are mostly product names.
Does the AI Overview fire on ecommerce queries?
In our own ecommerce set it fires on every single one. Across five capture rounds between 17 and 29 July 2026, we ran 21 ecommerce queries in four markets, and the AI Overview fired on 420 of 420 usable cells, with no genuine no-Overview cell anywhere in the study.
That sits next to Ahrefs’ 3.2% for shopping queries, and the two cannot both be describing the same thing. They are not. Ours is a small, deliberately built ladder; theirs is a huge panel. The gap between them is the whole argument of this article.
Here is our method, so you can judge it rather than take it:
- 21 queries, chosen to ladder competitiveness on purpose. The set deliberately includes near-zero-volume tail queries and AI-shopping questions, which is exactly where a blended average predicts the feature stops firing. It did not stop.
- Four markets, one row each: US as the anchor, then Canada, the UK and Australia. The market is asserted against Google’s own footer on every capture, never assumed.
- Signed out every time, desktop every time. A signed-in search is personalized, so it cannot tell you anything general.
- Two cells were blocked by Google on the first round and held back rather than counted. They were re-captured after a cooldown and both fired. Counting a block as a no-Overview is the easiest way to manufacture a false negative, so we do not do it.
The limits matter as much as the result. This is 21 queries in one vertical, on desktop, over two weeks, run by one operator. It is a counterexample to 3.2%, not a replacement for it, and I would not extend that 100% one query past the set we actually measured.
What it does establish is worth having anyway. On a real ecommerce keyword set in July 2026, the answer to “will an AI Overview even show up here” was yes, every time, including on the tail queries most people assume are safe.
Why am I not seeing AI Overviews on some searches?
Usually because of your market, your signed-in state, or the shape of the query itself. Presence is decided per query and per country, so the same search can return an AI Overview in one market and nothing at all in another on the same day.
We have watched exactly that happen for weeks. On the query mermaid diagram wordpress, the AI Overview has not fired in the United Kingdom for 13 consecutive capture rounds, while firing in the US, Canada and Australia on every one of those nights.
Being straight about how solid that is: nine of those 13 rounds were confirmed by opening the screenshot and reading the page, and four were confirmed by the capture tool’s own flag plus a file-size match to the established no-Overview pattern. Nine eye-checked rounds out of 13 is strong; claiming all 13 would not be true. It is also one query in one window, so it shows that coverage can be market-dependent, not that the UK gets fewer AI Overviews overall.
Before you conclude anything from your own spot checks, rule out the boring causes first. Sign out, because a personalized result tells you nothing general. Pin the market, because an unpinned search runs in whatever country Google thinks you are in. And check the same query on more than one day, because we have watched a citation appear and disappear on a page nobody touched.
If the Overview fires but your page is not in it, that is a different problem with a different diagnosis, and we walk through it in why your page is not cited in AI Overviews. If the page is not indexed at all, coverage rate is not your problem yet, and crawled, currently not indexed is the place to start.
How to measure your own AI Overview coverage rate
Take the queries you already rank for, search each one signed out with the market pinned, record whether an AI Overview fires, and read the rate off your own list. It is a genuine afternoon of work and it replaces every borrowed percentage in this article. We ran the same kind of first-party test to see whether an llms.txt file moves anything at all.
- Export your real query list from Search Console, not a keyword tool. Take the queries that already bring you impressions, because those are the searches where your visibility is actually at stake. Fifty is plenty to start; a hundred is better. If you are unsure which property to pull from, our note on Search Console platform properties covers the difference.
- Cut the list down to the intents that matter to you. Separate your question queries from your product and brand queries before you measure, because mixing them recreates the exact blended-average problem you are trying to escape.
- Search each query signed out, in an incognito or a clean browser profile. A signed-in search is personalized by your history, so it measures you rather than the search result.
- Pin the market you actually sell in. Country binds and city does not, so set the country and ignore the city line at the bottom of the page. If you sell in more than one country, run each one as its own list and never merge them.
- Record a plain yes or no per query, nothing clever. Whether the feature fires is a whole-page reading, which makes it the most reliable thing you can measure by hand. Whether you are cited inside it is much harder and needs a separate pass.
- Repeat the whole list on a second day, at least a week later. One sweep gives you a snapshot; two give you a sense of how much it moves. We have seen enough movement on unchanged pages to think one look is not enough.
- Divide, and write the denominator down next to the answer. “AI Overviews fired on 34 of our 50 highest-impression informational queries in the US in July” is a sentence you can act on. “AI Overviews appear on 43% of searches” is not.
That last step is the one people skip, and it is the one that makes the number usable. Without the denominator written down, your own figure decays into the same context-free statistic you started out mistrusting.
What an AI Overview coverage number cannot tell you
Coverage tells you whether the feature shows up, and that is all it tells you. It is a useful first reading precisely because it is cheap and unambiguous, but three questions sit outside it and they are the ones that decide what you should do.
- It does not tell you whether you lose clicks. An Overview firing above your result is not the same as it taking your traffic, and the effect varies by intent and by position. That is a measurement job of its own, and we set it out in how to measure AI Overview traffic loss.
- It does not tell you whether you are cited. Whether your page ends up inside the answer is decided separately from whether the answer appears, and we have watched a page hold a citation for four straight nights and then lose it without being edited. I wrote up one of those episodes in a piece on capturing the same AI Overview twice in one day, when two of its UK sources quietly vanished between captures.
- It does not cover mobile. Every capture behind our own two figures is desktop, and Google’s own UK answer says AI Overviews “show up most frequently on mobile devices”. If that is right, our desktop-only readings are a floor rather than a ceiling, and it is a real gap in what we can currently claim.
So, is a single AI Overview percentage worth anything?
As a headline for the industry, yes. As an input to your decisions, almost none. The published figures tell you the direction of travel, which is clearly upward and fast, and after that they stop being about you.
My honest position is that we keep trading one number that describes nobody, when the useful number takes an afternoon to produce and describes exactly the pages you are being paid to protect. We have a first-party 100% and a first-party 0% sitting in the same month as the 43% and 48% headlines, and the reason those coexist is not that anyone is lying. It is that “how often do AI Overviews appear” has no single answer until you say for whom.
So use the vendor studies for what they are genuinely good at, which is telling you which query types are exposed and which way the trend is moving. Then go and measure your own list, write the denominator next to it, and treat that as the only coverage figure allowed to change your plan.
Common questions about AI Overview coverage
Are AI Overviews still growing, or have they leveled off?
The published readings still point upward overall: Search Engine Roundtable covered this measure at about 13% to 20% a year ago and reports Similarweb at about 43% now. It has not been a clean line though, and at least one tracker recorded a pullback during late 2025, so treat the direction as reliable and the month-to-month wobble as noise.
Does a bigger keyword panel make a coverage study more accurate?
It makes it more stable, not more relevant to you. A 146-million-result study is an excellent description of that panel, and your forty pages are still not in it. Sample size fixes noise; it does not fix a denominator that was never about your site.
Can I trust what Google’s own AI Overview says about AI Overviews?
Treat it as a summary of the same vendor blog posts you could read yourself, not as a statement from Google Search. That is exactly why it returned a different range in two markets on the same day. It is a useful shortcut to what the published consensus looks like, and it is not a source of internal data.
Work with WpConsults
Want a hand measuring your own AI Overview exposure?
If you would rather not run the sweep yourself, send me your query list and I will tell you what your real coverage rate looks like and which intents are actually exposed.
Want our posts to show up more often on Google?
One step & Google will surface this site in your Top Stories.
