Generative Engine Optimization Statistics for Beginners

Generative engine optimization statistics are the published numbers about how ChatGPT, Perplexity, Gemini, Claude, Google AI Overviews and AI Mode use web content, and the headline "up to 40%" visibility figure from the GEO research paper comes from a 2024 lab benchmark, not from tracking real customers. That matters if someone is pitching you GEO work with that number. Re-checked against the authors' own released results, the top method moved a source's share of a lab-written answer from 19.4% to 27.3%. That's a real gain, and it's 7.9 percentage points, measured in a test setup rather than on the engines your customers open.

Key points

• The GEO numbers you'll see quoted most often come from lab research, engine usage counts or searcher behavior studies, and each kind answers a different question.
• Before you act on a figure, find who measured it, on which engine, when, and against what starting point.
• The only numbers that describe your business are the ones you collect yourself, by asking the engines your customers' questions and writing down what comes back.

What are generative engine optimization statistics?

Generative engine optimization statistics are figures that describe how AI engines find, use and cite web pages, or how people use those engines. They include research results on what makes a page more visible in an AI-written answer, user counts that engine makers publish, studies of how searchers click, and tracking data on which sites engines cite.

If the term itself is new to you, our plain GEO basics cover it. This page is about the numbers, so you can tell a useful one from a sales line.

Which generative engine optimization statistics come from primary sources?

Here are widely repeated figures, traced back to the organization that produced them.

Two notes on the usage rows. Both counts are reported by the companies themselves. And a weekly count and a monthly count can't be compared directly, because a monthly count includes anyone who used the product once in the month.

What does the "up to 40%" figure actually measure?

It measures how much of an AI-written answer was credited to one source page, before and after that page was rewritten. The researchers built a test set of questions, supplied each answer-writing model with a fixed set of sources, rewrote one source using different methods, and scored how much of the answer came from it.

Vector Index re-derived the figure from the authors' released results file, in an analysis last updated in October 2026. Across 989 queries, the unchanged source averaged 19.4% of the answer. Rewritten with the method called Quotation Addition, it averaged 27.3%. That's a 40.6% relative change, and it still scored below the unchanged page on 27.6% of the queries.

Two limits are worth knowing before anyone sells you on this number. The released code writes answers with gpt-3.5-turbo-16k from five sources supplied to it, so it never tests whether an engine would find your page in the first place. And the paper's separate Perplexity.ai test, where the best method on one measure showed a 22% improvement over the baseline, supplied the source text to Perplexity.ai as file uploads, so it doesn't show whether Perplexity would find your page on its own either.

Hypothetical example: an agency tells you GEO raises visibility 40%. If your page currently makes up 10% of an AI answer, a 40% relative lift takes it to 14%. That's 4 points, not 40. Ask any seller which kind of percentage they mean.

Do AI summaries change how people click?

In Pew's data, people clicked less on search pages that showed an AI summary. Pew tracked the browsing of 900 U.S. adults in March 2025, covering 68,879 unique Google searches. In that sample, 18% of searches produced an AI summary. Users clicked a traditional result in 8% of visits to those pages, compared with 15% on pages without a summary, and clicked a link inside the summary in just 1% of visits.

Pew recorded what people did. It didn't run an experiment, and summaries showed up far more often on some kinds of searches than others. Among searches of 10 words or more, 53% produced one. So the gap could reflect the kind of search as well as the summary, and the fair reading is that people clicked less on pages with an AI summary, not that the summary is proven to have caused it.

If your customers ask long, specific questions, many of them may read an answer before they ever see your link. Being named or cited inside that answer is one way to stay on their list.

How do you read a generative engine optimization statistic before you act on it?

Run any figure through these questions. If a page can't answer most of them, treat the number as a talking point.

1. Who measured it? A research team, the engine's maker, a survey firm or a company selling a tool.
2. Which engine? ChatGPT, Perplexity, Gemini, Claude, Google AI Overviews and AI Mode don't behave the same, and an average across them describes none of them.
3. When? Engines change often. A figure without a collection date can't be checked.
4. Compared with what? A relative change ("40% more") needs its starting point. Ask for the before and after.
5. What was counted? Words in an answer, citations, users, clicks and sales are different things. Visibility in an answer isn't a phone call.

Where do you get statistics about your own business?

You collect them. No figure above covers your trade in your city, and that's the number closest to whether a customer hears your name.

1. Write down the questions customers ask before they hire someone like you, in their words.
2. Ask each question in ChatGPT, Perplexity, Gemini and Google (where AI Overviews and AI Mode appear), with web search on where there's a choice.
3. Record the date, the engine, the businesses named and the pages cited. Note whether you appear at all.
4. Repeat on a set schedule with the same questions, so you're comparing like with like.

This costs time, not money, and it gives you a starting point for judging any GEO work later. If ChatGPT is where your customers ask, our checklist for ranking in ChatGPT covers what to fix first.

Before you pay for GEO work on the strength of generative engine optimization statistics, check the 40% figure for yourself. Vector Index's breakdown of the up to 40% figure, checked against its own baseline walks through the released results query by query, so you'll know exactly what to ask anyone who quotes it to you.

FAQ

Does generative engine optimization increase visibility by 40%?

The 2024 GEO paper says GEO "can boost visibility by up to 40%" in its benchmark. Re-derived from the released results, the top method moved a source's share of the answer from 19.4% to 27.3%, a 7.9 percentage point gain in a test setup, not a forecast for your site.

How many people use ChatGPT and Google AI Mode?

ChatGPT had 900 million weekly active users (OpenAI, February 2026). AI Mode passed 1 billion monthly active users (Google, July 2026). Both are company-reported, and weekly and monthly counts aren't directly comparable.

Do Google AI Overviews reduce clicks to websites?

Pew's March 2025 data shows fewer clicks on pages with an AI summary (8% of visits versus 15%). Pew observed the gap. It didn't prove the summary caused it.

Are there GEO statistics for my industry or city?

Not in the sources above. The GEO paper found that the effectiveness of different methods varies across domains, which is one more reason to measure your own questions on the engines your customers use.