Google AI Overviews Study: How Often They Appear by Query Type (October 2026)

Google showed an AI Overview on 77 percent of the US searches we checked. We fetched 494 Google results pages on desktop on October 9, 2026, and 380 of them had an AI answer on the page. The kind of search matters most: 97 percent of question searches had one, but only 34 percent of local searches did.

An AI Overview is the short answer Google writes with AI and places above the normal results, with links to the pages it used. This study counts how often it shows up, how many pages it cites, and which websites those are. You can download the full dataset and check every number.

Key findings

All numbers come from US, English language Google results fetched on October 9, 2026.

1. 77 percent of desktop results pages had an AI Overview. That was 380 of 494 searches.
2. Questions and comparisons almost always got one. 70 of 72 question searches (97 percent) and 68 of 70 “best” or “vs” searches (97 percent) had an AI Overview. Local searches had the fewest: 24 of 70 (34 percent).
3. Most cited pages did not rank in the top 10. The 380 AI Overviews cited 2,931 links. Only 776 of them (26 percent) were also top 10 organic results for the same search.
4. YouTube was the most cited site by far. It was cited in 204 of 380 AI Overviews (54 percent). Reddit was second, in 82 (22 percent).
5. Mobile and desktop showed AI Overviews at the same rate. On 154 keywords fetched on both devices, 114 had one on mobile and 113 on desktop. Still, 27 of those keywords had one on only one device.

How often AI Overviews appear by query type

We put every keyword into one of seven query types using fixed word rules. The rules are in the methods section. Then we counted the results pages that had an AI Overview.

The 95% range shows how far each share could move with a new sample of the same size. The local range runs from 24 to 46 percent, for example. So a gap of a few points between two types means little. The big gaps hold up: questions and comparisons sit far above brand and local searches.

SEOs call health and money topics YMYL, short for “Your Money or Your Life”. Bad advice on these topics can hurt people. Google still showed an AI Overview on 94 percent of them: 34 of 35 health searches and 32 of 35 money searches.

Local searches: the map pack still wins

Local searches had the fewest AI Overviews. The map pack with nearby businesses showed on 50 of the 70 local pages, and only 8 of those 50 also had an AI Overview.

The local group was mixed. None of the 7 dentist searches had an AI Overview. Movie and show searches had one 6 times out of 7, and hotel searches 5 times out of 7. When a local AI Overview did show, it often pointed to Google’s own business or hotel listings instead of websites. That happened on 11 of the 24.

A smaller sample we ran in September 2026 found no AI Overviews on 20 local searches. Two samples cannot tell us whether this is a change over time or just different keywords. One detail matters here, though: 15 of the 24 local AI Overviews in this study loaded after the rest of the page. A fetch that does not wait for late answers could miss most of them.

Brand searches: help and login searches get an answer

Searches for a brand or website were split. Brand searches with login, check in, customer service, support, download, portal, jobs, or careers had an AI Overview 18 times out of 19. The other 51 brand searches, mostly a business name alone, had one 23 times.

We did not plan that split before the study. Treat it as a lead to check on your own keywords, not a finding.

Which industries see the most AI Overviews

Read this table with care. Eight of the industries have the same mix of query types. Health and finance swap their questions and short topic terms for the 70 YMYL keywords. Most of the gap between industries comes from local and brand searches. Health, for example, drops because none of its dentist searches had an AI Overview.

Leave out local and brand searches and every industry but retail lands between 84 and 100 percent. Retail stays lowest at 22 of 32 (69 percent). On 8 of the 10 retail pages without an AI Overview, Google showed a grid of products with prices instead.

Longer searches got more AI Overviews

Searches of five words or more had an AI Overview 88 percent of the time (134 of 152). Searches of one to four words had one 72 percent of the time (246 of 342). Search volume made less difference: 70 percent for keywords with 100 to 999 monthly searches, 81 percent for 1,000 to 9,999, and 78 percent for 10,000 or more.

How many sources an AI Overview cites

An AI Overview cited 7.7 web pages on average. The middle value was 8. Most of them (277 of 380) cited between 5 and 11 pages.

Nine listed no sources at all. In six of them, the answer named local businesses and linked them to Google’s own place listings inside the text.

Health and money answers cited the most pages, 8.9 on average. Local answers cited the fewest, 4.8, and brand answers 5.5.

Do AI Overviews cite the pages that rank?

Mostly not the same pages. Of 2,931 cited links, 776 (26 percent) pointed to a page that was also in the organic top 10 for that search. At the site level the overlap was higher. About half of the cited links came from a website that had some page in the top 10.

So ranking well helps, but it does not get you cited by itself. Google’s documentation says AI Overviews may run several related searches and pull in more supporting pages while the answer is written. That may be why many sources come from outside the results you see. Google calls this query fan-out, and AI Mode uses it too, as we explain in AI Mode vs AI Overviews.

Price and buying searches had the most overlap: 35 percent of their cited links were in the top 10. Question, short topic, and YMYL searches had the least, at 22 percent each.

Which websites AI Overviews cite most

This table counts how many AI Overviews cited each site at least once.

Video stands out. YouTube links made up 484 of the 2,931 citations, or 17 percent. YouTube was cited in 48 of 70 question answers and 51 of 68 comparison answers.

Social and forum sites fill many of the next spots: Reddit, Facebook, Instagram, Quora, and TikTok. Behind them is a long tail of 1,291 different websites. Health answers leaned on Mayo Clinic, Cleveland Clinic, the American Heart Association, and the US National Library of Medicine. Money answers leaned on NerdWallet, Bankrate, and the federal student aid site. That last one ranks high partly because our money sample held many student loan searches.

How AI Overviews sit with other SERP features

A SERP feature is any block on the results page that is not a plain link, like People Also Ask or a map. Here is how often each one showed up next to an AI Overview.

Three patterns stand out.

• No featured snippets at all. Not one of 494 pages had one. Our featured snippet guide covers what replaced them.
• Forum and video blocks travel with AI Overviews. 50 of 51 pages with a forums block also had an AI Overview.
• Map packs and shopping grids mostly show without one. Only 31 percent of pages with a map pack had an AI Overview.

When an AI Overview showed, it was the very first block on the page 317 times out of 380 (83 percent). Across all 494 pages, the first organic result was the first thing on the page only 69 times (14 percent). Our guide to SERP features explains each block.

Mobile vs desktop

We fetched 154 of the keywords on a phone as well, 22 from each query type.

The rate was the same on both devices. 100 keywords had an AI Overview on both, and 27 had one on neither. The other 27 differed: 13 had one only on desktop and 14 only on mobile.

So about one keyword in six gave a different answer by device. When both devices showed one, they usually cited the same websites: 66 of those 100 pairs cited exactly the same set.

Mobile answers cited fewer top 10 pages. Part of that gap may come from the page itself, since mobile pages listed slightly fewer organic results (8.0 against 8.3 on average). For more on why the two pages differ, see why mobile and desktop rankings differ.

How we ran the study

The keyword sample

We pulled keywords from DataForSEO Labs, a keyword database, with its Keyword Ideas tool. Every keyword is a US English search with at least 100 searches a month, by the Labs estimate. We used 10 industries with 7 or 8 seed terms each, for example diabetes and dentist for health, or flights and hotels for travel.

We asked for each query type in its own request, using a word pattern or DataForSEO’s intent label as a filter. Six groups came back thin, so we ran one more request for each with extra seed terms. We did the same for the health and money groups, which leaned on too few topics. With one test request, that made 67 keyword requests.

Each keyword got exactly one query type. We checked these rules in order and used the first that matched:

Brand names that contain a rule word, like Best Buy, did not count as that word. We also removed near copies, such as two keywords with the same words in a different order, and kept the one with more searches.

Next we shuffled each industry and type pair with a fixed random seed and drew keywords in that order. We drew 7 per industry for local, brand, comparison, and price searches, so 70 each. We drew 9 per industry for questions and short topic terms from the 8 industries outside health and finance, so 72 each. The YMYL group took 35 health and 35 finance keywords.

We read every drawn keyword by hand. We removed 183 of the 677 we drew (27 percent) and replaced each with the next one in line:

That left 494 keywords: 70 or 72 per query type, across 10 industries.

How we fetched the pages

All 648 results pages were fetched through the DataForSEO SERP API on October 9, 2026, between 18:38 and 18:45 UTC. We used the Live Google Organic Advanced endpoint. The settings were United States (location code 2840), English, and the top 10 results. All 494 keywords ran on desktop, and 154 of them also ran on mobile.

Some AI Overviews load a moment after the rest of the page. We turned on DataForSEO’s load_async_ai_overview option so the fetch waits for them. That mattered: 114 of the 380 desktop AI Overviews (30 percent) loaded this way. For local searches it was 15 of 24.

Each fetch cost $0.002, plus $0.002 more when an AI Overview loaded late. The whole study cost $2.89 in API fees: $1.31 for keyword data and $1.58 for 648 results pages.

What we counted

• AI Overview present: the page held an AI Overview block. Three of the 380 blocks came back with no text. We still count them, because the block was on the page.
• Sources: the different web addresses the AI Overview linked to, from its source list and its inline links. Links back into Google itself, such as place cards and Google searches, are not counted as sources. 64 of the 380 AI Overviews had links like that.
• In the top 10: the cited address matched one of the first 10 organic results on the same page. We ignored http or https, www, a trailing slash, and anything after a # sign. A site match means the same website name.
• 95% ranges: Wilson score intervals for each share.
• Other features: People Also Ask, featured snippet, local map pack (hotel packs not counted), video or short video blocks, discussions and forums, and shopping or product grids.

Limits of this study

• One country, one day. AI Overviews change often, and other countries and languages will differ.
• A logged-out fetch. We saw what a fresh desktop browser or phone sees. A signed-in user can see a different page.
• Late-loading answers. We waited for AI Overviews that load late, but one that loads later still, or only after a click, would be missed. Our rates are more likely too low than too high.
• Keyword choice. The sample comes from one keyword database, our seed terms, and our rules. A quarter of the drawn keywords were removed by hand. The YMYL group leans toward blood pressure, diabetes, pregnancy, student loans, and mortgages.
• Equal weight per type. Each query type has about 70 keywords, whatever its share of real searches. So 77 percent is the rate for this sample, not for all Google searches.

That last point is one reason other studies report lower numbers. Pew Research studied the web browsing of 900 US adults in March 2025. An AI summary showed on 18 percent of their Google searches. Semrush found AI Overviews on 15.69 percent of queries in November 2025, across more than 10 million keywords. Both cover a broader mix of searches than our sample, and both are from an earlier date.

Download the data

Get the full dataset as a CSV file: google-ai-overviews-study-2026-10.csv. It has 648 rows, one per keyword and device: 494 desktop and 154 mobile.

The three AI Overview columns that only make sense when one showed are blank on pages without one. If you use the data, please link back to this page.

What this means for SEOs

Plan for an AI Overview on most informational keywords. If your keyword is a question, a comparison, or a health or money topic, expect an AI answer above you almost every time. The first organic result was the first thing on the page only 14 percent of the time.

Ranking and being cited are separate goals. Three out of four cited pages were not top 10 results for that search. Track both, because one does not guarantee the other.

Video and forums carry weight. YouTube was cited in over half of all AI Overviews, and Reddit in more than one in five. A short video or a helpful forum answer can be your way into the citations.

Local and brand searches still work the old way. The map pack and the brand’s own site still lead those pages. Your local rankings matter more there than AI citations.

How to track AI Overviews on your own keywords

Averages hide a lot. Your own keywords may sit at 30 percent or at 100 percent. Our guide on tracking AI Overviews and their citations shows the free checks, the Search Console report, and the tool options.

To repeat this study on your own list, use a SERP API that returns AI Overview blocks as data. Fetch each keyword for a fixed country and device, wait for late-loading answers, and save the source links. Then compare those links with your top 10, as we did here.