Generative engine optimisation

The disambiguation, because it costs the field real confusion

"GEO" already meant geographic. Geo-targeting, geolocation, geo-fencing. Some practitioners avoid the term entirely for that reason and use AEO, answer engine optimisation, instead. Both labels point at the same work. Rankability's analysis of 3,751 keywords across 48 months of Google Keyword Planner data found GEO carries roughly twice AEO's search volume, while practitioners converge on AEO as the standard label. There is no methodological difference between them worth arguing about.

This site uses GEO because that is its name, and covers both terms because the reader searching for one wants the other.

What the primary research establishes

Less than you would think from reading about it. The field's founding paper measured a research benchmark, not a commercial AI assistant, and its famous 40% figure has been carried a long way from what it actually says. The evidence that does concern live systems is mostly evidence about how unstable and fragmented they are.

Read those two together and a picture forms. Citation is drifting away from the pages that rank for your query, and it is fragmenting across engines that barely agree with each other. Both trends make a single "AI visibility" number less meaningful over time, not more.

What appears to move the needle

We are deliberately cautious here, because this is where the evidence thins and the confidence of the writing usually thickens. What follows is what large studies point at, with the honest caveat that almost all of them are correlational and several are published by companies selling tools.

• Being worth citing. Original data, named authors, falsifiable claims. Semrush's analysis of 11,882 prompts found promotional tone correlates negatively with citation.
• Being mentioned where the models already look. Earned media, communities and third-party publications recur across every study we examined. This is PR work, not on-page work.
• Being easy to extract. Answer the question in the opening sentences. Keep the number next to its source. Structure beats length.
• Being crawlable by the right agents. Unglamorous, but blocking GPTBot or OAI-SearchBot removes you from consideration entirely.

What does not appear to move it

• llms.txt. No study shows it helps. Google's John Mueller has said no AI service has confirmed using it. We could not reproduce the server-log figures usually cited against it either, so we cite neither. Graded.
• Generic schema markup. Any effect appears conditional at best, and mostly disappears once organic rank is controlled for.
• Buying a tracker. A founder who built one and sold it concluded it "didn't change what they needed to do". Graded.
• Length. The idea that long-form always wins is folklore. In the analyses we found, word count correlates near zero with citation.

Is it worth doing at all?

Yes, with the accent on the fundamentals rather than the acronym. ChatGPT reached roughly 900 million weekly active users by February 2026, and people use it to decide things. Being the source it reaches for is obviously worth having. What is not established is that this requires a separate discipline, a separate budget line and a $499-a-month subscription.

We have written the strongest available case against the whole field, using the words of the practitioners who make it best, on is GEO a scam?. If you are about to spend money on this, read that page before you read the ones selling you something.

Frequently asked

What is generative engine optimisation?

Generative engine optimisation, or GEO, is the practice of making a source more likely to be drawn on and cited when an AI assistant such as ChatGPT, Perplexity, Claude or Google AI Overviews generates an answer. The term comes from a 2023 academic paper by Aggarwal and colleagues, accepted to KDD 2024. It is unrelated to geographic or location-based targeting, which shares the abbreviation.

Is GEO the same as AEO?

They are used interchangeably in practice. GEO (generative engine optimisation) emphasises the generated answer; AEO (answer engine optimisation) emphasises the answer being given. Rankability analysis of 3,751 keywords over 48 months found GEO carries roughly twice AEO search volume, while practitioners increasingly converge on AEO as the standard label. No meaningful methodological difference separates them.

Is GEO different from SEO?

Partly, and less than the marketing suggests. The levers overlap heavily: authority, clarity, being mentioned in sources the models already trust. What has changed is the retrieval surface. Ahrefs found that the share of AI Overview citations drawn from pages ranking in the organic top 10 fell from about 76% in July 2025 to 37.9% in January 2026, which means optimising only for your own ranking pages addresses a shrinking share of citations.

Can I measure my GEO performance?

Not reliably yet. The same prompt returns the same list of brands fewer than 1 time in 100, and only 2.37% of cited URLs appear across all three major engines. Google Search Console AI performance report gives impressions with no clicks, click-through rate or queries. OpenAI publishes no analytics for ChatGPT at all. Treat any single AI visibility score as a weather vane rather than a speedometer.

Does llms.txt help?

There is no published study showing that it does. Evidence points the other way: AI crawlers appear largely to ignore the file, and Google John Mueller has said no AI service has confirmed using it. We could not reproduce the specific server-log figures most often cited against it, so we do not repeat them.