AI search statistics are easy to find and hard to trust. The same page can carry a figure with no source, a vendor's own client result, and a study with a stated sample, all in the same style.
Every number below comes from a page we opened, and the sentence or table names who measured it and how many searches, pages or people were in the sample. Where figures conflict, we say so. Where a figure came to us through a compilation, we say that too. Use the page to find numbers you can quote, and read the last section before you quote any of them. For action, see the AI SEO checklist and the AI visibility hub.
01How often AI Overviews appear
AI Overviews show on a minority of searches overall and on a majority of question searches. Ahrefs measured 146,122,391 desktop searches and found that 29,991,998 of them, or 20.5%, showed an AI Overview. The rate climbs with question words and query length, and falls on shopping and real estate.
Ahrefs also found AI Overviews on 57.9% of question queries against 15.5% of other queries, and on 24.9% of non-branded searches against 13.1% of branded ones. Seer Interactive, tracking 49,358 queries for 53 brands, found them on 95.4% of comparison queries and 85.9% of question queries, and on 36% of informational, about 8% of commercial and 5% of transactional queries. The two studies use different samples and definitions, so the rates differ.
| Category | Share of searches with an AI Overview |
|---|---|
| Science | 43.6% |
| Health | 43.0% |
| Medical (inside health and safety topics) | 44.1% |
| Pets and Animals | 36.8% |
| Finance | 24.9% |
| Law and Government | 23.1% |
| Travel and Transportation | 15.7% |
| Real Estate | 5.8% |
| Shopping | 3.2% |
Ahrefs, What triggers AI Overviews
02Who gets cited in AI Overviews
Most pages cited in AI Overviews already rank on page one. Ahrefs studied 1.9 million citations from 1 million AI Overviews and found that 76.1% of cited pages rank in Google's top 10, 9.5% rank between positions 11 and 100, and 14.4% rank below position 100. The median organic position across the three most visible citations was 3.
Brand mentions across the web matter more than links. In Ahrefs' study of 75,000 brands, branded web mentions correlated 0.664 with AI Overview mentions, against 0.218 for backlinks and 0.326 for Domain Rating. Brands in the top quartile for web mentions averaged 169 AI Overview mentions, against 14 for the next quartile. Correlation does not prove cause.
An older Semrush study of 200,000 keywords found that AI Overviews linked to an average of 11 sources and that only about 20% to 26% of those links overlapped with the organic top 10. Its sample predates the Ahrefs studies, and the overlap rose in the later, larger measurement. Read the gap as a change in the feature over time, not as a contradiction.
03How long cited pages are
Length has almost no link to being cited. Ahrefs analysed 174,048 cited pages and found 53.4% under 1,000 words, with a 0.04 correlation between word count and being cited. The average cited page ran 1,282 words, against 1,188 for the organic-ranking average.
| Page type | Median words |
|---|---|
| Listings (ecommerce, marketplace) | 315 |
| Core pages (home, about, service) | 317 |
| User-generated content | 387 |
| Video descriptions and transcripts | 407 |
| Interactive tools | 507 |
| Category pages | 534 |
| Documents (PDFs, slides) | 676 |
| Articles (blog posts) | 1,166 |
| Audio (mainly podcast transcripts) | 1,226 |
Ahrefs, Short vs long content in AI Overviews
Pew Research, studying browsing by 900 US adults, found that the median AI summary was 67 words, that 88% of AI summaries cited three or more sources, and that Wikipedia, YouTube and Reddit together made up 15% of sources in AI summaries against 17% in standard results. Government sites made up 6% of AI summary sources against 2% of standard results.
04What happens to clicks
Clicks fall when an AI answer appears, and the size of the fall depends on the study. Pew Research found that in visits to search pages with an AI summary, 8% included a click on a traditional result, against 15% without. About 1% included a click on a link inside the summary. Visits ended the browsing session 26% of the time with an AI summary against 16% without.
Ahrefs compared 300,000 keywords and found the top result's click-through rate was about 58% lower than forecast when an AI Overview appeared, with the effect shrinking down the page. This is correlation, in desktop data.
Seer Interactive's 53-brand data shows the other side. Organic click-through on queries that showed an AI Overview rose from 1.3% at its low to 2.4% two months later. Across the year, informational queries earned 2.07% organic click-through when the brand was cited in the AI Overview, 0.94% when it was not cited and 3.35% with no AI Overview. Seer calls this correlation and notes that higher-authority brands are also more likely to be cited.
| Situation | Approximate clicks |
|---|---|
| No AI Overview | 33,500 |
| AI Overview, brand cited | 20,743 |
| AI Overview, brand not cited | 9,445 |
Seer Interactive, AIO impact on Google CTR
A publishers' court filing reported Microsoft data in which click-through in Bing Chat was 51% to 94% lower than in Bing web search, depending on the publisher. It is one side's brief, the court has not ruled, and the dates and method are not disclosed. We include it as a reported figure, not a measured one.
05What an AI mention does to visits and revenue
A mention in an AI answer is followed by more brand visits, and most of those visits are hard to trace. Profound analysed more than two million US conversations across ChatGPT, Gemini and Google AI Overviews and found that brand site visits in the seven days after a mention ran about 1.5 to 2.5 times the forecast baseline: about 1.5 times for ChatGPT, 1.6 for Google AI Overviews and 2.5 for Gemini. For ChatGPT, 6.39% of mentions were followed by a visit against a forecast of 4.33%.
About 20.5% of first visits happened within an hour and 42% within 24 hours. More than 97% of post-mention ChatGPT brand visits carried no UTM tag, so analytics undercounts them. Profound sells AI visibility tools and the study is not randomised.
An Arcalea client case, covering a regional B2B distributor, found 198 orders worth about $116,000 with at least one trackable ChatGPT visit. About 49% of that revenue came from orders where ChatGPT was not the last touchpoint, and 72% to 84% of ChatGPT-assisted customers checked out as guests. It is one client's first-party analysis.
06The usage picture, with caveats
Usage figures vary widely by source, so we show the ones that are measured the same way. Suff Digital used Google Trends, a relative 0 to 100 index and not a count of searches. Over the last 52 weeks ChatGPT averaged 75.7, Gemini 5.8, Claude 4 and Perplexity 1. Against the 52 weeks before, searches for Claude rose about 347%, Gemini about 278% and ChatGPT 24.7%.
| Term | Average index | Change on the prior 52 weeks |
|---|---|---|
| ChatGPT | 75.7 | +24.7% |
| Gemini | 5.8 | about +278% |
| Claude | 4 | about +347% |
| Microsoft Copilot | 1.1 | slight rise |
| Perplexity | 1 | flat |
Suff Digital, AI Search Trends
Some usage numbers reach us only through compilations. Wix reports Perplexity as about 2% of AI search queries by its own research, and a Perplexity statement of 780 million monthly searches. AI Syndicate, citing Alphabet's earnings call, reports 2 billion monthly users of AI Overviews and more than 100 million monthly AI Mode users in the US and India, and cites a Gartner forecast of a 25% drop in traditional search volume. mean.ceo, citing Similarweb, reports AI referrals converting at 11.4% against 5.3% for organic. We did not open those original sources, so treat them as reported.
| Figure | Reported by | Original source named |
|---|---|---|
| AI Overviews reach 2 billion monthly users | AI Syndicate | Alphabet earnings call |
| Traditional search volume forecast to fall 25% | AI Syndicate | Gartner |
| Content with statistics and cited sources gets up to 40% more visibility | AI Syndicate and the GEO paper abstract | Aggarwal and colleagues, GEO study |
| AI referrals convert at 11.4% against 5.3% for organic | mean.ceo | Similarweb |
| About 60% of Perplexity citations overlap with Google's top 10 | Wix Studio | BrightEdge |
| Reddit is 46.7% of Perplexity's top 10 citations and 11.3% of ChatGPT's | Wix Studio | Profound |
07When AI gets facts wrong
One vendor, MaxAEO, counted 412 confirmed brand-fact errors across 63 B2B companies and found 61% traced to retrieval, 24% to training data and 15% to entity confusion. Crawler blocks were the sole cause in 7% of cases. Its median time to a corrected answer ran from 9 days for Perplexity to 33 days for Gemini. The vendor does not publish its method, so treat these as indicative. Our guide to fixing wrong ChatGPT information shows the diagnosis.
On the buyer side, the IAB's measurement framework says more than 20 companies sell AI visibility measurement tools, which is a reminder that the numbers you see depend on how a tool samples. Our guide to measuring AI visibility shows how to do it without relying on one score.
08How to read these numbers before you quote them
Check four things before you quote any statistic: who measured it, how many searches or people were in the sample, whether it shows correlation or cause, and whether the publisher sells something the number supports. mean.ceo's own compilation shows why. It reports an AI conversion premium of 2.15 times from one source, about 9 times from another and 23 times from a third, and flags the gap itself.
Before you quote an AI statistic
0 of 8 donePick the claim you need
For example, 'AI answers reduce clicks'. Write it as one sentence you want to support.
Find two studies with different methods
Pew uses browsing data, Ahrefs uses Search Console data. If both point the same way, the claim is safer.
Quote with the sample
Write the publisher, the sample and the number in one sentence.
Add the limit
One short clause on what the study cannot say, such as correlation only or desktop only.
Use this page if
- You need numbers with a named source and sample
- You are writing a report or a deck and want figures that survive checking
- You want to see where studies disagree
Look elsewhere if
- You need a forecast, since these are measurements and reported figures
- You need figures for a specific country or language, since most samples here are US or desktop
- You want a ranking of AI engines, which no source here supports
What most statistics pages give you
- A long list with no sample sizes
- Vendor figures mixed with measured ones
- Conflicting numbers with no comment
- Figures with the year in the sentence, which dates fast
- A call to buy a tool at the end
What this page does
- Every figure traced to a page we opened, named in the sentence
- Samples stated beside the number
- Conflicts shown and explained
- Secondary-source figures set apart in their own table
- A path to action in our Perplexity guide and Google AI Mode guide
