Generative engine optimization statistics: which numbers are real, and which to ignore

Only a handful of the numbers quoted about GEO lead back to something you can open and check. Even those describe one benchmark, one browsing panel or one survey at one point in time, so use them as context for your own measurements and never as a forecast for your site. Nobody can guarantee that a page will be included in an AI answer, and no statistic changes that.

If you are new to the topic, start with what GEO is and how it differs from SEO.

How to grade a GEO statistic before you quote it

Ask five questions of any figure. Who published it? Can you open the study itself? What was the sample, and who was in it? What date, platform and country does it cover? What do the authors say it cannot show?

If you cannot answer the first two, stop there. A round percentage with no named study is a claim. It might be true, but neither you nor your reader can check it.

Then put it in one of three grades:

Grade	What it is	What it can support	What it cannot
Platform documentation	The company that runs the product describes how it works	How the system behaves and how it is reported	Outcomes for any particular site
Named research	A paper or analysis with a method you can read	What happened under the stated conditions	Other countries, periods, engines or sites
Vendor or consultancy claim	A figure from someone who sells services, tools or advice	That the publisher says it	Anything, until the method is shown

What platforms themselves document

Google’s page “AI features and your website” is the clearest platform statement. To be shown as a supporting link in AI Overviews or AI Mode, “a page must be indexed and eligible to be shown in Google Search with a snippet”, and it adds: “There are no additional technical requirements.” It also says “specific optimization isn’t required for AI Overviews and AI Mode”, while normal SEO fundamentals “continue to be worthwhile”.

On measurement, the same page says sites appearing in AI features “are included in the overall search traffic” in Search Console, reported in the Performance report under the “Web” search type. Search Console’s help page adds the counting rules:

• Clicking a link to an external page in an AI Overview or AI Mode counts as a click.
• An AI Overview occupies a single position, and every link in it is given that same position.
• A follow-up question in AI Mode is treated as a new query.
• Data from Search Labs experiments is not included.

Neither page describes a separate report or filter for AI feature traffic, so you cannot read an “AI Overviews share” straight out of Search Console from what Google documents.

One line on Google’s page deserves a grade of its own. Google says that clicks from results pages with AI Overviews “are higher quality (meaning, users are more likely to spend more time on the site)”. It publishes no figure or method for this, so treat it as the platform’s own claim about its own product.

We found no number in OpenAI, Perplexity or Anthropic crawler documentation that bears on visibility statistics, so they do not appear here.

What named research has measured

The “up to 40%” paper

Aggarwal et al., “GEO: Generative Engine Optimization” (accepted to KDD 2024), is where the term and the most quoted number come from. What the paper actually did:

• Benchmark. GEO-bench, 10,000 queries drawn from nine sources, about 80% of them informational.
• Engine. For each query, the authors fetched the top five Google results and had gpt-3.5-turbo write an answer citing them, sampling five answers per query.
• Methods. Nine content rewrites, including Cite Sources, Quotation Addition, Statistics Addition, Fluency Optimization and Keyword Stuffing.
• Metrics. Position-Adjusted Word Count (how much of the answer is attributed to a source, weighted by where its citation appears) and Subjective Impression (seven factors scored by GPT-3.5).

The best methods improved on the unchanged baseline by 41% on Position-Adjusted Word Count and 28% on Subjective Impression. The authors summarise this as “up to 40%”. Keyword stuffing gave little to no improvement, and results varied by domain. They also tested the methods on Perplexity.ai and report improvements of up to 37% there.

The limits matter more than the headline. The gain is a relative change in one source’s share of an answer built from five sources, measured mostly on an engine the authors built themselves with a 2023 to 2024 model. It does not measure traffic, rankings, leads or sales. Quote it as “a research benchmark reported relative visibility gains of up to 40% under its own conditions”.

Pew Research Center on clicks

Pew’s 2025 analysis is independent research with a published method. It used the browsing data of 900 US adults in its KnowledgePanel Digital panel, who agreed to have their browsing tracked, from 1 to 31 March 2025. That gave 68,879 unique Google searches, of which 12,593 showed an AI summary. Pew re-ran the searches with a scraping service between 7 and 17 April 2025 to see what each results page contained.

What it found:

• Users clicked a traditional search result on 8% of visits to pages with an AI summary, against 15% of visits without one.
• They clicked a link inside the AI summary on 1% of visits to pages that had one.
• They ended their browsing session on 26% of pages with an AI summary, against 16% without.
• 18% of the Google searches in the study produced an AI summary.

Its limits: Google only, US adults only, one month, and results pages were collected a few weeks after the searches were made, so some summaries may have changed. A “click” was inferred from the next URL visited.

Why vendor and roundup numbers need care

A company that sells GEO services, tools or advice has a reason to publish a big number. That does not make the number false, but it comes with an interest attached and the method is often missing.

Roundups make it worse. One page quotes a percentage, the next quotes the first page, and soon a figure appears on dozens of sites with the original sample, date and wording stripped away. By then “80% of search users rely on AI summaries at least 40% of the time” has often become “80% of people use AI to search”.

Common headline claims and what is behind them

We traced the four figures that turn up most often in the current top results.

Claim as usually quoted	Who actually published it	What we could check	Grade
“80% rely on AI summaries”	Bain & Company press release, 19 February 2025: “About 80% of search users rely on AI summaries at least 40% of the time”	The press release gives no sample size or method on the page. It also claims this reduces organic web traffic “by 15% to 25%”	Consultancy claim
“60% of searches end without a click”	The same Bain release: “On traditional search engines, about 60% of searches now end without the user progressing to another destination”	No method on the page	Consultancy claim
“Organic traffic down 50% by 2028”	A Gartner Insights abstract (5 May 2026): “Gartner predicts organic search traffic will decrease by 50% by 2028”	It is a forecast, and the research behind it is paywalled, so we could not read the method or assumptions	Analyst forecast
“37% start searches with AI”	Eight Oh Two, an SEO and PPC agency, as reported by Search Engine Land	A survey of 500 consumers who actively use AI tools, data collected in November. The full report requires registration, so we read the coverage only	Agency claim

Two points stand out. The 37% describes people who already use AI tools, so it says nothing about consumers in general. And it has nothing to do with the “up to 37%” Perplexity.ai result in the Aggarwal paper; the two numbers simply share a value.

None of this makes the figures wrong. It means each one is the publisher’s own claim, and should be quoted with that publisher’s name attached.

What you can measure for your own site

This section is our recommendation, not a research finding. Your own numbers will tell you more than anyone’s headline, because they describe your pages and your buyers.

1. Search Console for Google. Watch clicks and impressions in the Performance report. Remember that AI Overview and AI Mode activity is counted inside the “Web” search type rather than reported separately.
2. Referrals in your analytics. Look for visits referred by AI assistants and note how they change over time.
3. A fixed set of buyer questions, checked on a schedule. Write down the questions your customers actually ask, run them on the platforms you care about, and log whether your business is named, cited and described accurately.

Keep the question set fixed so the comparison means something, and record the date, platform, location and exact wording each time. Do the same for any outside statistic you quote: write down its sample size, date, platform, prompt set and location. More on setting this up is in the measurement section of our guide and in what AI visibility means. For Google specifically, see how to appear in AI Overviews.

How to use statistics in a business case without overclaiming

Lead with what you measured yourself. Add outside figures as context, each labelled by grade.

Some wording that stays accurate:

• “A KDD 2024 research benchmark reported relative visibility gains of up to 40% for some content changes, under its own test conditions.”
• “Pew Research Center found that in March 2025, US adults in its panel clicked a traditional result on 8% of Google visits with an AI summary, against 15% without.”
• “Bain & Company says about 80% of search users rely on AI summaries at least 40% of the time.”

Never turn any of these into a promise. A sound business case says what you will track, how often, and what you will do if the numbers do not move.