SEO, GEO and AEO: how to get your brand cited in AI answers

Three names, one job: be the source an answer is built from

Buyers now ask a question and get a written answer, from Google AI Overviews, ChatGPT search, Perplexity or Claude. The links under that answer are the new first page. Three terms describe the work of getting there, and they overlap more than the acronyms suggest.

In practice they are one programme: technical SEO decides whether you can be found at all, AEO decides whether your words can be quoted, and GEO decides whether you are the source worth quoting.

• SEO (search engine optimization): gets a page crawled, indexed and ranked in classic results and snippets. It rewards relevance, links and technical health.
• AEO (answer engine optimization): shapes a page so its answer can be lifted into featured answers, voice replies and AI Overviews. It rewards a clear answer in the first lines.
• GEO (generative engine optimization): makes a page a source that ChatGPT, Perplexity, Claude, Gemini or AI Mode cite when they write an answer. It rewards credible, quotable, sourced facts.

What Google says about AI Overviews and AI Mode

Google is direct about it: "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary." To be shown as a supporting link, a page has to be indexed and eligible to appear in Search with a snippet^[1].

Google also says you do not need new machine-readable files, AI text files or special schema.org markup for these features, and that structured data should match the visible text on the page^[1].

Two consequences follow. Technical health still decides everything: a page blocked from Googlebot or marked nosnippet cannot be a supporting link. And the controls are the familiar ones: nosnippet, data-nosnippet, max-snippet and noindex limit what Search shows, while Google-Extended governs AI training and grounding in some of Google's other systems, not Search itself^[1].

Measurement is already in place: traffic from AI Overviews and AI Mode is counted in the Search Console Performance report under the Web search type^[1].

What the GEO research found

The term GEO comes from a 2023 paper by Pranjal Aggarwal and colleagues, accepted at the KDD 2024 conference. The authors built a benchmark of real queries and tested nine ways of rewriting a source page to see which ones made generative engines show it more^[2].

Read the headline number carefully: 30 to 40% is a relative improvement on a research benchmark, not a promise for your site. Use it as a direction (evidence beats adjectives), then measure on your own pages.

• The best methods, citing sources, adding quotations and adding statistics, improved visibility by 30 to 40% relative to the unedited page^[2].
• Classic keyword stuffing did not help in generative answers^[2].
• Results varied by domain, so the authors recommend testing per category rather than one recipe for every page^[2].

A checklist to get cited

Work through these in order; the first four change how a page reads, the rest change how machines find and trust it.

At Datakult the Keyword Master agent runs this list inside the Keyword and Demand Intelligence sprint: it maps the questions buyers ask in each language and drafts answer-first sections, and a person approves every change before it goes live.

• Answer first: put a 40 to 60 word answer to the page's main question in the opening lines, in plain words a model can quote.
• One question per section: phrase headings the way buyers ask, and answer each one in its first sentence.
• Bring evidence: add numbers you can source, quotes from named experts and links to primary documents, with dates^[2].
• Show who stands behind it: a named author, published and updated dates, and Google's "Who, How and Why" questions answered on the page^[3].
• Mark up what is on the page: Article, FAQ and Organization structured data in JSON-LD that matches the visible text^{[1, 4]}.
• Keep entities consistent: the same company name, description and facts across your site, profiles and directories.
• Let the right crawlers in: allow the search user agents of the assistants you want to appear in, and decide separately about training crawlers^{[5, 6, 7]}.
• Tell search engines when pages change: sitemaps with real modification dates, and IndexNow for Bing, Yandex and the other engines that support it^[8].
• Optional: an llms.txt file. It is a proposal rather than a standard, and Google says it does not need one; some AI tools read it^{[1, 9]}.

Which AI crawlers to allow

Each assistant publishes separate user agents for search, for visits a user triggers and for model training, and each one obeys its own robots.txt rule, so the choices are independent^{[5, 6, 7]}.

A sensible default for most brands: allow the search and user agents so you can be found and cited, and treat training crawlers as a separate, deliberate business decision.

• OAI-SearchBot (OpenAI): shows sites in ChatGPT search results^[5].
• ChatGPT-User (OpenAI): visits a page when a user asks ChatGPT^[5].
• GPTBot (OpenAI): collects content that may be used to train models^[5].
• PerplexityBot (Perplexity): surfaces and links sites in Perplexity search and is not used to train foundation models^[6].
• Perplexity-User (Perplexity): visits a page when a user asks Perplexity^[6].
• Claude-SearchBot (Anthropic): improves search result quality for Claude users^[7].
• Claude-User (Anthropic): visits a page when a user asks Claude^[7].
• ClaudeBot (Anthropic): collects content that may be used to train models^[7].

How to measure AI visibility without inventing numbers

You do not need a vendor score you cannot check. Four simple measures give you a baseline in a month and a trend in a quarter.

• Google: AI Overviews and AI Mode clicks are inside the Search Console Web performance report, so watch pages and queries rather than a separate AI line^[1].
• Assistants: track sessions referred by the assistants' domains in your analytics, and the conversions they bring.
• Citations: keep a fixed panel of 20 to 50 buyer questions per market and language, ask them monthly in each assistant, and record whether and how you are cited.
• Content: log which pages gained an answer-first summary, sources or structured data, so you compare like with like.