Published estimates of how often Google shows an AI Overview run from about two in five searches to more than four in five. They cannot all describe the same thing. This piece sets out the main published estimates, our reading of why they disagree, and what a South African business should measure instead of quoting a headline percentage.
The estimates
All of these are published in 2026:
- 43%: Similarweb, as reported by Search Engine Roundtable on 30 July, up from about 15%. The report does not give the start date for that comparison. The same report puts Semrush at about 48%.
- Over 62%: seoClarity, quoted by Search Engine Roundtable on 26 October as covering queries in general, with a range of 55% to 84% depending on industry.
- Over 80%: a blended claim in the same Search Engine Roundtable report, drawn from Semrush, SE Ranking and seoClarity data, up from 50% in June 2025 and about 15% the year before that. The only per-source figures in that article are SE Ranking’s 82.5% of US brand searches and seoClarity’s over 62% above, so the blend is the reporter’s summary, not one measurement.
- 86.7%: Peec AI, a vendor, from 500,000 prompts run between 13 and 20 April 2026. Peec’s figure for April 2025 was 56.9%.
Our own GEO and AI statistics page currently carries a Semrush figure of 25.8% for United States searches in January 2026. That was a fair snapshot then, and it now reads as stale next to the figures above. We are refreshing that page to show the range, not a single number.
Why they disagree
What follows is our reasoning, not a finding from any one study.
The keyword set decides the answer. Peec’s own breakdown shows how much. AI Overviews appeared on 64.6% of two-word prompts and up to 89.1% of prompts of 11 to 15 words, with an average of 87%. A study that samples short keywords will report a lower share than one that samples long, conversational ones.
Country and sector matter. Peec’s figure for the EU was 76%, and seoClarity’s industry range runs from 55% to 84%. A blended number hides both.
The share has been moving quickly. A 25.8% reading in January and an over-80% reading in October are not necessarily in conflict. Some of the spread between estimates is real growth, and some is method, and a single article cannot tell you which.
Collection method matters, and recent data may be exposed to a second problem. Nozzle, a rank-tracking vendor, reports that Google has been returning altered results pages to tracking tools since September. Search Engine Journal, relaying Nozzle’s data, says AI Overviews appeared in 34.9% of altered pages against 54.2% at baseline. If that holds, any prevalence figure collected by a tracker during September and October may understate the true share. The April and July figures above predate September, so the effect would apply only to autumn collection. We cannot say how much any particular estimate was affected, and Google has made no statement, so this is a question to ask a data provider, not a conclusion. The background is in our piece on altered results.
What a prevalence figure does not tell you
Even a perfectly measured percentage answers only one question: how often an AI Overview is shown. It does not say whether your brand is mentioned, whether you are linked, or whether anyone clicks. Those are three separate outcomes, and they move independently. A page can rank well and never be cited, which is the subject of our piece on why page-one rankings do not buy a seat in the AI answer. So when a headline says AI Overviews now appear on most searches, the useful follow-up is not “how many” but “on which of my searches, and what happened to my visits when they did”.
What it means for a South African business
Algorithm’s October 2026 study of clients’ sites checked 5,850 non-branded page-one terms and found an AI Overview on 4,100 of them, or 70%. By funnel stage the share was 75.4% for learning searches, 74.5% for buying, 77.0% for comparing and 34.9% for navigating.
Two cautions apply. It is an agency analysing its own clients, and each term was checked once. But the pattern is the useful part: AI Overviews are least common where someone is trying to reach a known site, and most common where someone is learning, buying or comparing, which is where a business is trying to be discovered. We cover the click data from the same study in South African businesses are being seen more and clicked less.
What to measure on your own queries
- List your 20 to 40 searches that matter. Use the questions buyers ask and the comparisons they make, not only your brand name.
- Record what appears, monthly. For each query, note whether an AI Overview shows, whether your brand is mentioned and whether you are linked. Run the check the same way every time, with the same device and location.
- Match it to Search Console. Look at impressions, clicks and average position for those queries in the same months. If a query gains an AI Overview and its clicks fall while position holds, you have evidence about your own business.
- Use ranges when you quote the market. Say “published estimates run from 43% to over 80% depending on the source” and cite them.
- Ask any provider how its prevalence data is collected, and whether it checks for altered results.