What is generative engine optimization (GEO)?

In plain terms, GEO asks one question of every important page: could a system that writes answers instead of listing links fetch this page, make sense of it and name it as a source? Most of the work is ordinary good publishing, and so far the formal research behind the term is one benchmark study.

The short definition, and what it covers

Wikipedia describes GEO as “the practice of improving visibility in responses generated by generative artificial intelligence (AI) systems”. We use that to check how the word is generally used. It is not evidence that any tactic works.

In practice GEO covers three things:

• Access: whether AI crawlers, and the search indexes behind AI answers, can reach and read your pages.
• Content: what you say, how plainly you say it, and whether each claim can be checked.
• Consistency: whether your business is described the same way wherever it appears.

It does not cover buying a place in an answer. None of the platform documentation we read for this page describes a paid route into AI answers.

Every claim below is one of three kinds, and we say which: a platform’s documented behaviour, a named research finding with its limits, or our own recommendation. Where we had no source, we left the point out. The home guide has a shorter version of this definition in what is GEO.

Where the term came from

The term comes from a paper: Aggarwal et al., “GEO: Generative Engine Optimization”, first posted to arXiv in November 2023 and accepted to KDD 2024. The authors present GEO as a way to help content creators make their content more visible in generative engine responses. They also released GEO-bench, a set of 10,000 queries across several domains, for testing methods.

The word has since spread well beyond academia, and plenty of agencies use it loosely. When someone quotes a GEO result at you, ask whether it traces back to something you can read.

How AI answers are produced, and why that matters for GEO

Different systems work differently, so we only describe what the platforms themselves document.

Google AI Overviews and AI Mode are built on Google Search. Google’s documentation says that to be shown as a supporting link, “a page must be indexed and eligible to be shown in Google Search with a snippet”. It also says “There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary”. According to the same page, there is no special schema.org structured data to add and no need for new machine readable or AI text files. Both features may use “query fan-out”, which means running several related searches across subtopics to build one response. Google also says plainly that “Indexing and serving isn’t guaranteed.”

OpenAI, Perplexity and Anthropic each document separate crawlers for separate jobs:

Company	Crawler	What the company says it is for
OpenAI	OAI-SearchBot	Showing websites in ChatGPT’s search features
OpenAI	GPTBot	Crawling content that may be used to train its foundation models
OpenAI	ChatGPT-User	Visiting pages when a user asks; OpenAI says robots.txt rules may not apply
Perplexity	PerplexityBot	Showing and linking websites in Perplexity’s search results; not used to train foundation models
Perplexity	Perplexity-User	Fetching a page when a user asks; Perplexity says it generally ignores robots.txt
Anthropic	ClaudeBot	Collecting web content that could contribute to model training
Anthropic	Claude-SearchBot	Improving the quality of Claude’s search results
Anthropic	Claude-User	Fetching pages when a Claude user asks a question

The practical point is that search and training are controlled separately. OpenAI, for example, says sites that opt out of OAI-SearchBot are left out of ChatGPT search answers. Blocking GPTBot only signals that your content should not be used for training. If you block every AI bot by reflex, you may take yourself out of AI search results as well.

Access is the first gate. If a system cannot fetch or index your page, it cannot cite it, and everything else in GEO depends on that. The crawler access section covers the robots.txt detail, and how it works gives the wider picture.

What GEO involves in practice

This section is our recommendation. No platform confirms that these steps lead to citations.

1. Put clear, direct answers near the top of the page. If a question has a short answer, give it first, then explain. This helps readers whatever the AI systems do; we cannot prove that answer systems favour it.
2. Name your sources and make claims checkable. Say where a number came from and link to it. If you cannot point to evidence, soften the claim or drop it.
3. Keep important text crawlable. Words that only appear after scripts run may not be seen by every crawler. Plain text in the page is the safer choice.
4. Describe your business consistently across the web. Use the same name, description and facts wherever you appear.

Most of this is good editorial practice under a new label, which is why it is worth doing even as the AI products change. The principles section goes through each one in more depth.

Which crawlers you let in is your decision, and robots.txt is how you state it. The standard, RFC 9309, says its rules “are not a form of access authorization”; crawlers are “requested to honor” them. Anthropic says its bots honour robots.txt directives. OpenAI and Perplexity say their user-triggered fetchers may not follow it. Decide which bots to allow based on each company’s own description of what they do.

What research has and has not shown

The Aggarwal et al. paper is the main named study, so here are the conditions it ran under.

• The engines: the main tests used the authors’ own simulated engine, which took the top five Google results for a query and had gpt-3.5-turbo write the answer. They also tested Perplexity.ai on 200 of the queries. Because Perplexity does not let users choose source URLs, they uploaded the source text as files.
• The methods: nine ways of rewriting a source page: Authoritative, Statistics Addition, Keyword Stuffing, Cite Sources, Quotation Addition, Easy-to-Understand, Fluency Optimization, Unique Words and Technical Terms.
• The measures: two visibility metrics the authors designed themselves. One is a word count weighted by where a source is cited in the answer. The other is a “subjective impression” score covering seven aspects.

Under those conditions, the paper names Cite Sources, Quotation Addition and Statistics Addition as its top-performing methods. It reports that keyword stuffing offered little to no improvement. The abstract’s headline figure is a visibility boost of “up to 40%” on those metrics, and it notes that how well the methods work “varies across domains”.

The authors say methods may need to change as generative engines evolve, and that the questions people ask change over time. We would add two cautions. This is one study, mostly run on a simulated engine built in 2023 and 2024, and its gains are measured on its own metrics. A result from that setup does not automatically carry over to ChatGPT, Gemini or AI Overviews as they work today. Treat it as early evidence: worth reading, but not a rulebook.

What GEO cannot do

It cannot guarantee you appear in an answer. That is our stance, and the documentation supports it. Google describes eligibility and says serving is not guaranteed, and none of the crawler documentation we read promises inclusion. Being eligible is not the same as being chosen.

It cannot make an engine behave predictably, because you do not control the engine. It cannot make up for having nothing worth citing. And nobody outside these companies can tell you for certain how a given model picks its sources, because the platforms do not publish that.

If an agency promises a guaranteed spot in ChatGPT or AI Overviews, ask them for the platform documentation that backs it.

GEO, AEO, LLMO and AI SEO: the names

These labels overlap. Wikipedia lists answer engine optimisation (AEO), large language model optimisation (LLMO) and “AI SEO” as other terms for the same concept. In our reading, different writers use them for similar and sometimes identical practices, and no standards body defines any of them.

Our working view is that GEO is the only one of these terms with a named research origin, while the others are common in industry writing. For how GEO relates to traditional search work, see GEO vs SEO. We will compare AEO in detail on a separate page.

Where to start

A short, honest list, in this order:

1. Check that your key pages are indexed and can show a snippet in Google Search. That is Google’s stated requirement for its AI features.
2. Read each AI company’s crawler documentation and decide, bot by bot, what to allow in robots.txt.
3. Rewrite your most important pages so the answer comes first and the sources are named.
4. Make sure the important text is in the page itself, not only rendered by scripts.
5. Make sure your business is described the same way on your site, profiles and listings.

The full 30-day version is in getting started. To check whether any of it shows up in AI answers, see measuring progress.