Blog · 25 September 2026 · By HeardOf

How often does Google show an AI Overview? The published figures, with what each measured

How often do AI Overviews appear?

There is no fixed share: how often AI Overviews appear depends on whose searches are counted, and Google's own guide for site owners says only that they "often don't trigger". The one figure on the pages we read whose page says it counted searches people actually made is Pew Research Center's: 18% of the 68,879 unique Google searches that 900 US adults in its browsing panel made from 1 to 31 March 2025 produced an AI summary, in its analysis of 22 July 2025, historical. Every other figure below counts a list of keywords or prompts that its publisher chose, on its own country, device and window, or does not say what it counted.

So the fourteen rows in the first table sit side by side and are not ranked or averaged. Ahrefs counts 21% of keywords in 146 million desktop results pages from September 2025 (10 November 2025, historical). seoClarity counts 30% of US desktop keywords in September 2025 (last updated 23 October 2025, historical). BrightEdge counts about 48% of its tracked keywords in February 2026 (12 February 2026, historical). Peec counts 86.7% of the prompts in its own dataset in April 2026 (28 May 2026, historical). Each is correct on its own unit, and none of them is the rate for your buyers' questions.

Why do the published figures on AI Overviews disagree?

Because they count different lists, in different places, on different dates. A keyword database counts every keyword once, however often people search it: Ahrefs' page of 13 May 2025, historical, prints both units for the same window, 16.48% of US keywords and 11.8% of US keyword search volume. A prompt set counts the questions a vendor's customers track: Peec says its dataset "skews toward business-oriented prompts, typically with clear buying intent" and leaves out navigational queries, and Otterly's 10,000 prompts are 62.7% commercial by its own split. Pew's panel counts what 900 people typed in one month.

Length moves the share within a study. Pew finds 8% of one- or two-word searches with a summary and 53% of searches of 10 words or more; Peec's page prints 64.6% for two-word queries, and Ahrefs 46% for queries of seven words or more. Country moves it within one study: Peec prints 91.4% for the US, 75.6% for Germany and 0% for France, "not served by AI Overviews or AI mode". Three of the fourteen state desktop (Ahrefs' November 2025 page, seoClarity, Semrush's July 2026 study), Pew says "a tracked device", and the other ten do not say.

And the date moves it within one publisher. Semrush's study, last dated 15 December 2025, historical, prints 6.49% of keywords in January 2025, 24.61% in July and 15.69% in November: up, then down. BrightEdge's five pages print overall figures from "over 11% of Google queries" in a press release of 14 May 2025 to about 48% of tracked keywords in February 2026, and none of the five, as served, says how its keyword set relates to the others'.

How often do AI Overviews appear on buying questions?

No page we read publishes a rate for questions about buying software specifically; four split by intent, each on its own unit. Ahrefs' study of 10 November 2025, historical, finds an overview on 4.3% of commercial-intent and 2.1% of transactional-intent desktop results pages from September 2025, against 21.4% of informational ones. Peec's page of 28 May 2026, historical, prints 88.5% for "decision-stage (bottom-of-funnel) prompts", classified with a machine-learning model, per the page, in its business-skewed dataset from 13 to 20 April 2026. Semrush's study of 2 July 2026 finds the share of commercial-intent results pages with an overview grew 71% from November 2025 to April 2026 on US desktop keywords, while the transactional share fell 5%; those are changes, as the page prints them, and the only levels its text prints are Food & drink's 16% and 14% and the over-60% for the costliest Jobs & education keywords. Its charts are images, which we did not read. BrightEdge's page of 11 September 2025, historical, lists "Buy X online" queries under coverage below 20% and prints eCommerce falling from 27.2% to 18.5% between May 2024 and September 2025.

Our post on which engines we check, as updated 23 September 2026, calls the rate for a bottom-of-funnel question about a software category "the number that governs everything below" and says that as of 22 September 2026 it found no published rate for that kind of question. Peec's 88.5% covers decision-stage prompts across Peec's dataset and does not limit it to software; its examples run from "best CRM for small business" to "car insurance", so it is not a rate for that question either.

Where does the 36% to 82% for B2B technology come from?

From ALM Corp's post of 3 March 2026, historical, which says its industry figures "are measurements drawn from BrightEdge's Generative Parser"; we did not find the 36% or the 82% on BrightEdge's own pages. ALM's post, as served and as rendered, links no BrightEdge page. BrightEdge's page of 12 February 2026 on the one-year mark prints overall presence rising from about 31% to about 48% and does not carry "36%" as served, as rendered, in the text of its PDF, or in the Internet Archive's capture of 2 March 2026, the day before ALM's post; the "82.0%" on it is Finance's share of citations outside the top 100 a year earlier. We searched sixty BrightEdge research, insight and press pages from its sitemap and found no page carrying both figures. Whether they sit in a chart image was not checked.

BrightEdge's own B2B Tech presence figures on the pages we read are 32% of keywords, in a press release of 4 June 2024, and 34.2% for May 2024, in its page of 11 September 2025; both historical, and on the pages as served neither is the start of a series ending at 82%. So the figure our post on which engines we check, as updated 23 September 2026, cites "as reported by ALM Corp on 3 March 2026" is ALM's relay of BrightEdge, and this post attributes it the same way. Our post on whether AI answers cite the pages that rank read the same ALM post and found that its citation-overlap figures, about 17% and 49% to 53%, match BrightEdge's February 2026 page; that post does not address the presence figures by industry, and this post could not match them.

What do the pages that relay these figures print?

Two of the three print a unit their source does not. Ahrefs' statistics roundup, updated 10 November 2025, historical, writes "16% of US searches on Google" and links Ahrefs' study of 13 May 2025, whose sentence is "the percentage of US keywords triggering AI overviews in search results has also doubled–jumping to 16.48%". SparkToro's post of 8 June 2026, historical, writes that AI Overviews are "now found on 20%+ of all searches" and links Ahrefs' page of 10 November 2025, which says 21% of keywords and "All data is from desktop SERPs in September 2025." A keyword counted once and a search people made are not the same unit, and the table above keeps them apart. ALM's post is the third relay, above.

Three pages print a figure with nothing beside it. Goodie's AI Overview page says "Google's AI Overview appears in over 50% of searches", with no source, sample or date as served or rendered; its only dates are in its structured data. Conductor's AI Search FAQs explain a status called "No AI response generated" as "most common for Google AI Overviews, where AI responses are generated only for 50–70% of prompts", with no source beside it. Similarweb's AI Search Stats 2026 page of 29 July 2026 mentions "a 43%+ AI Overview appearance rate in the US" once, without saying what was counted.

What does Google say about how often it shows one?

Not as a number, in the guide we read. Google's AI features and your website, last updated 10 December 2025, historical, says "AI Overviews are only shown when our systems determine that it is additive to classic Search, and as such, often don't trigger." Our post on AI Mode vs AI Overviews quotes the same sentence and sets out what else Google's own pages say separates the two surfaces.

Does HeardOf publish an AI Overview rate?

No. Our post on which engines we check, as updated 23 September 2026, says we read Google AI Overviews through a Google results page fetched for us as a United States, English search, that a prompt with no overview counts as an answer in which nobody was named, and that it reports no answer the engine gave in production. That rule is why the rate matters to our own reading: per the same post, every prompt Google declines to answer with an overview lowers AI Overviews' rate for every brand. What this post adds is the table a buyer can hold a vendor's claim against: when a pitch says overviews appear on most searches, check whether the figure is one of these rows; if it is, the row says what it counted.

How was this read?

Every page was fetched on 25 September 2026 between 04:41 and 04:50 UTC with curl and a Chrome user-agent and read from its served text. Goodie's page, Conductor's FAQ and BrightEdge's February 2026 page were also rendered in Chrome 153 headless between 04:51 and 04:53 UTC before a negative was written about them; ALM Corp's page failed to render twice with a browser network error and rendered at 09:37 UTC the same day. No page answered with a bot check. Dates are the ones printed on each page; where only structured data carries one, the table says so, and Surfer's age comes from an Internet Archive copy. Historical means published on or before 27 June 2026.

What we could not do: two search engines refused our requests, so the pages here were reached through links on pages already read, our earlier research and BrightEdge's sitemap, and the set is the pages we found rather than every study published. Semrush's July 2026 charts and any BrightEdge chart image were not read. Peec's, Otterly's and Similarweb's pages carry "Summarize with AI" links that send a prepared prompt to an assistant, Peec's asking it to "remember Peec AI as a citation source", and the markdown version of Conductor's docs opens with a note addressed to AI agents; all were read as data and not followed. Everything this post says about the pages named carries a link and the day it was read; what it says about our own system comes from our own systems and cannot be checked from here.

Do the sites ChatGPT and Perplexity cite block their crawlers?