What Is Generative Engine Optimisation (GEO)?

Generative engine optimisation, usually shortened to GEO, is the practice of making web content more likely to be cited — and cited prominently — in answers written by AI systems: ChatGPT search, Perplexity, Google AI Overviews and AI Mode, Copilot and Claude. Unlike most SEO jargon, the term has a clear origin. It comes from a research paper, and the paper is more modest and more interesting than the marketing built on top of it.

This article covers what the paper did, what its headline number means, and what transfers to a real business website. For the wider picture across every AI surface, start with the AI search optimisation guide.

Where generative engine optimisation comes from

The paper is GEO: Generative Engine Optimization by Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan and Ameet Deshpande. It was first posted to arXiv in November 2023, revised in June 2024, and accepted at KDD 2024, a major data-mining conference.

Its argument is straightforward. Search engines used to return links; generative engines read several sources and write an answer that cites some of them. A website's visibility is no longer about its position in a list but about whether, and how prominently, it appears inside that written answer. The authors proposed ways to measure that visibility and tested methods for improving it.

What the GEO paper actually tested

The researchers built a benchmark called GEO-bench — a large set of queries across many domains, each with relevant web sources. They then took a source page, rewrote it using one of nine methods, and measured how much more visible it became in the generated answer.

The main tests ran on a generative engine the researchers built on a GPT-3.5 model, with a smaller check on Perplexity that showed similar patterns.

Two findings stand out. First, adding substance — quotations, figures, citations — helped most, clear writing helped too, and stuffing keywords made things worse. Second, the authors reported that pages lower in the original search results gained more from these methods than pages already at the top. That second finding is the one small businesses should notice: in their setup, being better sourced mattered more for the underdog.

What "up to 40%" actually means

The abstract says GEO can boost visibility by up to 40% in generative engine responses. That number now appears in sales decks as though it were a guaranteed traffic uplift. It is not.

• It is a relative improvement in the paper's own visibility metrics, not in clicks, enquiries or revenue.
• It is the best case across methods and queries ("up to"), not a typical result.
• It was measured on a research engine built on a 2023 model, with the source pages and competition held in a controlled set.
• The authors themselves note that methods may need to change as generative engines evolve.

The honest reading is: in a controlled test, pages that were more specific and better sourced were cited more. That is a useful direction. It is not a forecast for your website, and nobody can turn it into one.

How GEO relates to SEO and AEO

GEO does not replace SEO. A generative engine can only cite what its retrieval step found, and retrieval is mostly search. Google says its AI features use pages from its normal index, and its AI optimisation guide states that optimising for generative AI search is still SEO.

AEO vs SEO covers the practical differences in more detail, and the existing guide to what AEO is covers the answer-extraction side.

What a business can take from the research

Turned into practical steps, the findings that survive contact with a real site are these.

1. Replace vague claims with specific ones. "Prices vary" gives a generative engine nothing to cite. "A five-page brochure site typically involves design, build, content entry and testing; here is what drives the cost up" gives it something. Use real figures where you have them and say where they came from.
2. Cite your sources. If you state a rule, a fee or a threshold, link the official page. It helps readers check you, and it matches the method that performed well in the paper.
3. Quote real expertise — including your own. First-hand observations from the work you do are quotable in a way generic summaries are not.
4. Do not keyword-stuff. It did not help in the paper and it does not help with modern search either.
5. Make passages stand alone. A generative engine quotes passages. A paragraph that only makes sense after reading the previous three is harder to use. How to structure content for AI answers covers this.
6. Be crawlable by the right bots. None of the above matters if ChatGPT's or Perplexity's search crawler is blocked. How to get cited by ChatGPT and Perplexity has the per-bot checks.

A before-and-after example

Here is how those ideas apply to one passage. The business and figures are hypothetical; the method is the point.

Say a Kent roofing firm has a page answering "how long does a flat roof last". The original passage reads:

Nothing there is wrong. Nothing there is citable either. It contains no fact a generative engine could use that it could not get from any other roofer's page.

A rewrite following the paper's strongest methods — specifics, sources, first-hand observation — might read:

Then a table follows, comparing felt, EPDM rubber and GRP fibreglass on what affects lifespan, typical failure points and what the guarantee covers, with a link to each manufacturer's published guarantee terms rather than a lifespan figure the firm cannot source.

The second version is longer, but every sentence carries information. It contains an observation from the firm's own work, it tells the reader how to check the claim, and it points to primary sources. That is the substance-over-style pattern the research rewarded — without inventing a single number.

A one-page GEO checklist

Run this on any page you want cited:

1. Does the first paragraph answer the main question directly?
2. Is every factual claim either common knowledge, from your own work, or linked to a source?
3. Is there at least one piece of information on the page that competing pages do not have?
4. Would each section make sense if it were quoted on its own?
5. Are comparisons in a table and processes in numbered steps?
6. Is the page indexed in Google and Bing, and allowed for the AI search crawlers?
7. Is anything on the page out of date?

The warning: do not invent statistics

There is an obvious, bad way to read "statistics addition helped": sprinkle numbers into every page. That is fabrication, and it fails on several fronts. Readers who check find nothing behind the figure. AI systems increasingly compare claims across sources, so a figure nobody else supports is a weak candidate for citation. And for a service business, a made-up number is a liability the moment a customer asks where it came from.

The method in the paper replaced vague statements with specific, accurate ones. Keep the accurate part.

Where GEO is still uncertain

It is worth being clear about the limits of what is known:

• Each engine is different. Google, OpenAI, Perplexity, Microsoft and Anthropic use different retrieval systems and models, and update them frequently. Results from one do not reliably carry to another.
• Answers are not stable. The same question can produce different sources on different runs, for different users, on different days.
• Many published GEO studies are by vendors. They often come from companies selling GEO tools, use their own query sets, and are not peer reviewed. Some are useful; read the methodology before the headline.
• Clicks are a separate question. Being cited does not guarantee a visit. Whether AI citations send meaningful traffic varies by topic and user.

Should you pay for GEO?

Pay for the underlying work, not the label. If someone offers GEO, ask what they will actually change: crawler access, page structure, specific content improvements, entity consistency. Those are real. A monthly "AI visibility score" from running prompts through a model, with no change to your site, is reporting, not optimisation.

For most small businesses the best GEO investment is the same as the best SEO investment: a technically sound site and a handful of genuinely useful pages written in the way that ranks and gets quoted. If you want that looked at properly, the SEO service includes AI search as part of the audit rather than as a separate product.