What is GEO? Being found and cited in AI answers

More and more questions get answered by an AI assistant before anyone clicks a link. Someone asks ChatGPT for a supplier, Perplexity for a comparison or Google for advice, and gets a written answer with a handful of sources. GEO, generative engine optimisation, is the work of making sure your content is found, used and cited in those answers.

Where the term comes from

The term comes from a research paper, GEO: Generative Engine Optimization by Pranjal Aggarwal and colleagues, first posted in November 2023 and presented at the KDD conference in 2024. The authors describe generative engines as systems that retrieve information and use large language models to compose an answer, and note that content creators have little control over how their content shows up in them.

They built a benchmark of queries, GEO-bench, and tested nine ways to rewrite a web page. Citing credible sources, adding quotations and adding statistics worked best, with visibility gains of up to 40 percent in their tests. Keyword stuffing, an old SEO reflex, brought little or nothing. Results differed per domain, and pages that ranked lower gained the most. Keep in mind that this is a lab study on the engines of that time: a direction, not a guarantee.

How AI answers pick their sources

When an assistant cites sources, it has usually searched first and then written an answer based on what it found. Google calls this grounding, or retrieval-augmented generation, and explains that its AI features also run several related searches at once, which it calls query fan-out. Those features are rooted in the same ranking and quality systems as regular Google Search.

Each assistant has its own crawler for this. OpenAI uses OAI-SearchBot to show sites in ChatGPT search, Anthropic uses Claude-SearchBot to improve search results in Claude, and Perplexity uses PerplexityBot. The Gemini app can ground its answers in Google’s Search index, which site owners can limit with a robots.txt token called Google-Extended, without affecting Google Search itself. Microsoft cites sources in Copilot and in AI summaries in Bing. The practical consequence is simple: a page that a crawler cannot reach cannot be cited.

The same companies use separate crawlers to collect training data, such as GPTBot at OpenAI and ClaudeBot at Anthropic. So you can decide separately whether your content may be used for training and whether it may appear in answers.

GEO and SEO: overlap and difference

The overlap is large. Google states that SEO best practices remain relevant for its AI features, that a page must be indexed and eligible to show with a snippet to appear as a supporting link, and that no special markup or AI text files are required. From Google’s point of view, optimising for generative AI search is still SEO.

The difference lies in what success looks like. In classic search, you win a position and, with luck, a click. In an AI answer, you win a mention, a citation or a sentence that carries your point of view, often next to competitors and sometimes without a click. And there is no single engine: ChatGPT, Claude, Gemini, Perplexity and Copilot each search and cite in their own way.

In an AI answer you do not win a position. You win a mention, a citation or a sentence.

What you can measure

• Google Search Console has a generative AI performance report, rolled out to all websites by the end of August 2026. It shows impressions in AI Overviews, AI Mode and AI features in Discover, by page, country, device and date.
• Bing Webmaster Tools has had an AI Performance report in public preview since February 2026. It shows how often your pages are cited in Copilot, in AI summaries in Bing and in some partner integrations, with a sample of the queries behind those citations.
• Your analytics: ChatGPT adds utm_source=chatgpt.com to the links it sends, so that traffic shows up as its own source.
• Your own test set: a fixed list of questions your customers really ask, put to several assistants at regular intervals. Note whether you appear, how you are described and which sources are cited instead.

Be wary of tools that claim to use Google’s internal metrics: Google says no third-party tool has access to its ranking or AI systems.

What you can do

1. Let the right crawlers in. Check your robots.txt for the search crawlers above, and decide separately about the training crawlers. In Google Search Console, the Search generative AI setting includes sites by default; an excluded site gets no impressions or traffic from these features.
2. Write what only you can write. Google asks for non-commodity content: first-hand experience, your own data, a clear point of view. A summary of what is already online adds little.
3. Back up your claims. Cite sources, quote experts, add figures. That is what worked best in the GEO paper, and Bing gives the same advice.
4. Make answers easy to find. Clear headings and sections, short direct answers to the questions people ask, and important content in text rather than only in images.
5. Keep facts current. Google and Bing both point to their business listings, such as Google Business Profile, Merchant Center and Bing Places, for local and product details in AI answers.
6. Skip the hacks. Google says that for its Search you can ignore llms.txt files, chopping content into small chunks and chasing inauthentic mentions. Producing pages for every variation of a question, mainly to steer AI answers, even breaks its spam policy.

GEO is something you can measure. In Zaela, a question like “How visible are we in ChatGPT and Gemini?” returns how often, where and how AI assistants mention you, compared with your competitors, with the questions behind it.