Is it true that GEO increases AI visibility by 40%?

Where does the 40% figure come from?

From the first academic GEO paper, published in 2023 by researchers at Princeton and IIT Delhi. If the term is new to you, start with our explainer on what GEO is.

Aggarwal and colleagues (opens in a new tab) wrote that their methods “can boost visibility by up to 40% in generative engine responses.” Their engine was a simulation. For each question, only the top 5 Google results were fetched, and GPT-3.5 wrote an answer citing them.

The team then rewrote one of those five sources in nine ways, such as adding statistics, quotations or citations, and compared it with the original. The best methods improved the page’s share by 41% on one measure and 28% on another. That 41% is the source of the rounded “40%” quoted in sales decks.

What exactly went up by 40%?

A page’s share of the words in an AI answer, among five pages already handed to the AI.

Martinez’s 2026 review (opens in a new tab), a critical survey of the field, spells it out. The headline comes from quotation addition, which lifted the page’s share from 19.3 to 27.2, or about 41% in relative terms. The share is weighted toward sentences near the top of the answer. The five pages’ shares always add up to 100, so one page can only gain what the others lose.

The review is blunt about what this does not mean. The result does not mean that 40% more readers will click, nor that a page will gain 40% in its chance of being retrieved. It means that a source already in front of the AI received a larger share of the text.

The gains were also uneven. When sources were cited more, the page that started fifth gained 115.1%, while the page that started first lost 30.3%. The method mostly helped pages the search engine had ranked low.

Did the result hold on a real AI search engine?

Partly, in a limited test on Perplexity that still supplied the pages directly.

The 2023 paper also tried its methods on Perplexity, a live AI search engine. The largest gains there were 22% and 37% on the two measures. But the review notes the test used 200 examples, and the texts were uploaded as files rather than found on the open web. It shows Perplexity can use a better-written page more, not that Perplexity will find it. How an old search trick fared in that test is covered in our guide to keyword stuffing in AI search.

Have later studies confirmed it?

Only within lab setups; tests closer to real search found much smaller, inconsistent or even negative effects.

Puerto and colleagues (opens in a new tab) built a benchmark across six kinds of content, from news to retail products, and four AI models. In their main experiment, out of 54 combinations of tactic and content type, only three gave a reliable gain, and many pushed pages down. Moving a page higher in the list the AI received beat every rewrite. When more competitors used the same tactic, the gains shrank toward zero. We review those tactics one by one in our guide to statistics, quotes and citations.

Newer methods learn their own rewriting rules. One of them, AutoGEO from Wu and colleagues (opens in a new tab) at Carnegie Mellon, reported an average improvement of 35.99% in GEO measures. That gain, the review notes, was again measured on five documents already retrieved.

The review also describes a 2026 preprint by Kim and colleagues that put search back into the loop, using a large pool of documents. Rewriting only a page’s body text cut its presence in the reranked top 10 by about 16% and its final citations by about 6%. A rewrite can win inside the answer and still lose the race to be found.

Does GEO raise traffic or sales?

The evidence is very thin. The review rates any link between citation scores and clicks or revenue as “very low” confidence.

One study the review covers examined a website’s logs after some pages were reworked for AI answers. ChatGPT referrals to those pages grew 5.7 times. But untreated pages on the same site had already grown 3.5 times as ChatGPT itself grew. The authors’ estimate of the extra lift was suggestive, not proven.

A separate study of 112 startups by Sharma (opens in a new tab), a single-author paper from IIT Patna, scored their websites on GEO features with a simple automated check. The score showed no link with whether ChatGPT or Perplexity recommended them in discovery questions. Referring links from other sites mattered on Perplexity instead.

Our own AI Overview study makes the same distinction. Among 3,096 pages ranking in Google’s top 10, ranking position explained more of which pages were cited than all 14 page features together.

What should you do about it?

Treat “40% more AI visibility” as a lab result, and ask vendors who quote it what they will measure.

1. Ask which measure a promised gain refers to: being found, being cited, being cited first, or traffic and sales. These are different results.
2. Ask for a baseline and a comparison group. A rise in AI referrals can come from the AI product growing, as the log study showed.
3. Expect results to vary by assistant, question and day, and insist on repeated measurement across several assistants.
4. Keep paying for the basics that the research supports: relevant pages that answer buyer questions, real evidence, and strong search ranking.
5. Be wary of fixed recipes. The review found no technique with a lasting effect on being found across several AI engines. See the GEO practices research supports instead.

If you want a measurement plan built this way, see our generative engine optimization service.

What does the research not tell us yet?

The field has strong lab evidence and almost no proof of business outcomes.

• No reviewed study shows a stable, long-term effect of any GEO technique on being found across several AI engines.
• Almost no studies measure clicks, leads or revenue with a proper comparison group.
• Most experiments fix the set of pages in advance, so they skip the step where most pages fail: being retrieved at all.
• AI products change often, so a result from 2023 or 2025 may not hold today.
• Gains measured with one page optimizing may disappear when competitors do the same.

Frequently asked questions

Where does the 40% GEO statistic come from?

It comes from Aggarwal and colleagues’ 2023 paper that named GEO. In a simulated engine built on GPT-3.5, adding quotations to one of five sources raised its share of the answer’s text by about 41%.

Does GEO guarantee more traffic from ChatGPT?

No. No published study shows that, and the 40% result measured share of an answer’s words, not clicks. In the one log study reviewed, reworked pages grew 5.7 times while untreated pages grew 3.5 times, so much of the rise came from ChatGPT’s own growth.

Is GEO worth investing in if the 40% is overstated?

It can be, if you judge it on measured results rather than the headline. The best-supported work is helping AI systems find and use relevant, evidence-rich pages, measured across several assistants over time.

Has anyone replicated the 40% GEO result?

Not outside similar lab setups. A 2025 benchmark found reliable gains in only three of 54 tests, and a test that included search found rewrites reduced final citations by about 6%.

Sources

• Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande (2024), GEO: Generative Engine Optimization (opens in a new tab), arXiv:2311.09735.
• Martinez (2026), Optimizing Visibility in Generative Engines: A Critical Survey of Generative Engine Optimization (2023-2026) (opens in a new tab), arXiv:2607.14035.
• Puerto, Gubri, Green, Oh and Yun (2025), C-SEO Bench: Does Conversational SEO Work? (opens in a new tab), arXiv:2506.11097.
• Wu, Zhong, Kim and Xiong (2025), What Generative Search Engines Like and How to Optimize Web Content Cooperatively (opens in a new tab), arXiv:2510.11438.
• Sharma (2026), The Discovery Gap: How Product Hunt Startups Vanish in LLM Organic Discovery Queries (opens in a new tab), arXiv:2601.00912.
• Underneath (2026), What pages cited by AI Overviews have in common: 3,096 pages