What is generative engine optimization (GEO)?
How AI assistants choose which sources to cite, and what you can change.
On this page
Short answer
This glossary explains the words that come up when a business starts working on visibility in AI answers: in ChatGPT, Perplexity, Gemini and Google’s AI features. Each entry says what the term means in practice and what, if anything, you can do about it on your own site. Terms are grouped by topic, so related ideas sit next to each other.
The disciplines
Generative engine optimization (GEO)
Work aimed at getting a brand named and its pages cited when an AI assistant writes an answer. The term comes from a 2023 research paper by authors from Princeton, IIT Delhi and two independent researchers, who tested which page changes made sources appear more often in generated answers. Full guide: what GEO is.
Answer engine optimization (AEO)
An older name for preparing pages so that a system can lift a direct answer from them: featured snippets, voice assistants, and now AI answers. In everyday use AEO and GEO overlap heavily. Full guide: what AEO is.
LLM SEO
A loose label for the same family of work, with the stress on large language models: what they learned about your company and what they find when they search. Full guide: what LLM SEO is.
AI SEO
Used in two senses: SEO done with the help of AI tools, or SEO aimed at AI search. When an agency says “AI SEO”, ask which one they mean.
How AI answers are built
Large language model (LLM)
The engine inside ChatGPT, Gemini, Claude and similar assistants. It composes a reply word by word, choosing likely continuations from patterns it absorbed during training.
Training data
The text a model learned from before release. What it “knows” about your company from training is frozen at that point and can be out of date or wrong.
Retrieval and grounding
Some assistants look things up in a search index or database first and write the reply from the results. Such an answer is called grounded. That is why a page published last week can show up in an answer today, while the model’s own training may be a year old. Researchers call the approach retrieval-augmented generation (RAG), introduced in a 2020 paper by Lewis and colleagues, then at Facebook AI Research.
Query fan-out
Google’s name for a technique used in AI Overviews and AI Mode: instead of one search, the system issues several related searches across subtopics and sources, then assembles the answer from what comes back. A page can therefore be linked for a sub-question the buyer never typed, such as “price range” or “how long it takes”. Google describes it in its AI features guide.
Hallucination
A confident statement from a model that is not true, such as a wrong price, a service you do not offer or an office that does not exist. Clear facts on your own site and consistent profiles elsewhere give the assistant something correct to retrieve.
Google’s AI features
AI Overviews
A short AI-written answer that Google places above the regular results for some searches, with links to sources. It reached US users in May 2024. It does not show for every query, and its presence for a given query can change over time.
AI Mode
A Google search mode built around follow-up questions: you talk to the search, and each reply is an AI answer with links. Like AI Overviews, it can use query fan-out.
Zero-click search
A search that ends without a click to any website, because the answer is already on the results page. Where an AI answer covers a simple question, fewer people may need to click, so a link inside that answer becomes one of the few ways to be seen.
Search Console and AI features
According to Google, traffic from AI Overviews and AI Mode is counted in the normal Search Console performance report, under the “Web” search type. There is no separate AI view, so judge the effect by comparing pages and queries over time.
Used in two senses: SEO done with the help of AI tools, or SEO aimed at AI search.
Being seen in answers
Mention
The assistant names your brand in its answer, with or without a link.
Citation
The assistant links to one of your pages as a source. A citation can come without a mention, for example when your guide explains a term but a competitor is recommended.
AI visibility
How often you are mentioned or cited across a fixed set of buyer questions, assistants and dates. Because answers vary from run to run, it is measured over repeated runs, not from one screenshot. You can check yours for free.
Share of voice in AI answers
Your mentions as a share of all brand mentions in the same set of answers. Useful for comparing yourself with three or four named competitors.
Telling machines who you are
Entity
A thing a search engine or model can identify as one specific object: your company, a person, a place, a product. Consistent name, address and description across your site and outside profiles help systems treat them as one entity.
Knowledge Graph
Google’s database of entities and the links between them. A knowledge panel on the right of search results is one visible sign that Google recognises an entity.
Structured data (schema.org)
Code in a page that states facts in a fixed format: organization name, address, prices, questions and answers, authors. It helps search engines read a page correctly. It does not guarantee any special display.
E-E-A-T
Experience, Expertise, Authoritativeness and Trust. Google’s search quality rater guidelines use these to tell human raters how to judge a page, with trust as the central one. Google does not treat it as one ranking score, yet it is a useful checklist for citable pages: a named author, first-hand experience, visible sources.
Crawlers and access
AI crawlers
Bots that fetch pages on behalf of AI companies, usually with separate bots for training, search and user requests. OpenAI lists GPTBot (training data), OAI-SearchBot (ChatGPT search results) and ChatGPT-User (visits a page when a user’s request calls for it; OpenAI says robots.txt rules may not apply to it). Perplexity lists PerplexityBot (search results, not model training) and Perplexity-User (user-triggered visits, which generally ignore robots.txt). Anthropic lists ClaudeBot (training), Claude-SearchBot (search quality) and Claude-User (user questions). Blocking a search bot can keep your pages out of that assistant’s answers. Sources: OpenAI, Perplexity, Anthropic.
Google-Extended
A robots.txt token that lets a site decide whether content Google crawls may train future Gemini models or be used for grounding in them. Google states it does not affect inclusion or ranking in Google Search. Pages stay in Search, and AI Overviews are part of Search, so this token is not the switch for them.
llms.txt
A proposed text file at the root of a site that lists its most useful pages for language models. Jeremy Howard of Answer.AI put the idea forward in 2024. Google says you do not need special AI text files to appear in its AI features, and we know of no major assistant that has confirmed using the file. Cheap to add, low priority.
robots.txt
The file that tells bots which parts of a site they may crawl. For AI visibility the usual mistake is a rule copied from somewhere that blocks search bots of AI assistants along with training bots.
The assistant names your brand in its answer, with or without a link.
Related terms from classic SEO
Hreflang
A tag that tells Google which language or country each version of a page is meant for. See international SEO.
White label
Work done by one agency and sold by another under its own brand. See what white label SEO is.
Featured snippet
A short answer box Google shows above regular results, taken from one page. It came before AI Overviews and still appears for some queries.
