This glossary defines the words that come up in AI search work: GEO, AEO, AI SEO and LLM SEO, the crawlers that fetch your pages, and the metrics that tell you whether AI models name your brand. Every entry opens with a one or two sentence definition, then adds the detail that matters when you have to act on it.
The terms are grouped rather than alphabetical, because most of them only make sense next to each other. If the whole field is new to you, read What is GEO? A practical guide first and come back here for the vocabulary.
Key takeaways
- GEO, AEO, AI SEO and LLM SEO describe largely the same work: getting AI models to find, understand and cite your brand.
- Three families of AI bots exist, and confusing them in robots.txt is the costliest mistake: search bots, user triggered fetchers, and training bots or tokens.
- Visibility in AI answers is a rate, not a position, because the same prompt can produce different answers on different runs.
- A citation is a link, a mention is your name, and share of voice is your name next to your competitors. The three are not interchangeable.
The four acronyms: GEO, AEO, AI SEO and LLM SEO
GEO (Generative Engine Optimization)
The practice of optimizing your content and online presence so that generative engines such as ChatGPT, Gemini, Claude and Perplexity mention and cite your brand in their answers. The term was introduced by the 2024 research paper GEO: Generative Engine Optimization, which frames it as a black box optimization framework for improving content visibility in generative engine responses, reports up to 40% more visibility on its benchmark, and stresses that the effect of each tactic varies by domain.
AEO (Answer Engine Optimization)
Content shaped so that a machine can lift a direct answer out of it. The label is older than GEO and dates from the era of featured snippets, voice assistants and question boxes, when the "answer engine" was still a search engine. In current usage the two overlap almost completely: AEO emphasizes the answer format, GEO emphasizes being chosen as a source by a generative model.
AI SEO
An umbrella label used in two different ways, which is why it causes so much confusion in briefs: SEO work aimed at AI surfaces, or using AI tools to do ordinary SEO faster. Before you agree on a scope or a budget, ask which of the two the other person means.
LLM SEO
The same idea as GEO with the model named instead of the engine. There is no formal definition and no separate methodology behind it, so treat it as a synonym and pick one term for your team.
None of these acronyms replaces classic SEO. AI answers are built on search infrastructure, and Google is explicit that the best practices for SEO remain relevant because its generative AI features are rooted in its core Search ranking and quality systems (Google Search Central). What carries over and what genuinely changes is the subject of GEO vs SEO: what changes and what stays the same.
Where AI answers appear
Generative engine
Any system that answers a question in prose instead of returning a list of links: ChatGPT, Gemini, Claude and Perplexity are the ones most brands are measured in. The answer names only a handful of brands, which is the structural reason AI visibility is harsher than a ranking: there is no second page.
AI search
The mode in which an assistant searches the web before answering, reads the pages it finds, and normally links them under or inside the answer. This is where most linked citations come from, and it is the part of AI visibility you can influence in weeks.
AI Overviews and AI Mode
Google's generative AI features inside Google Search. Google documents them together and gives no separate optimization guidance for them: the features draw on its core Search ranking and quality systems, structured data is not required for them, there is no special schema.org markup to add, and you do not need to write in a specific way just for generative AI search (Google Search Central). In practice they are a reason to keep your SEO fundamentals healthy rather than a separate project.
How a page reaches an AI answer
Retrieval
The step in which the model turns a question into searches, pulls candidate pages from an index and reads them before writing the answer. It is also called retrieval augmented generation. Pages that are easy to fetch and easy to quote win here, which is why retrieval is the fast lane of AI visibility.
Training memory
What a model absorbed about your brand while it was trained, also called parametric knowledge. It lets a model describe and recommend you without looking anything up, and it moves on the slow clock of model updates rather than on the clock of your publishing calendar.
Grounding
Attaching a generated answer to sources fetched at answer time so it stays factual. Google's Google-Extended token is the clearest documented example: it controls whether crawled content may be used for training future Gemini models and for grounding in Gemini Apps, and Google states that it does not affect a site's inclusion in Google Search and is not used as a ranking signal there.
AI crawler
Any automated fetcher run by an AI provider. Three families exist, they serve different purposes, and blocking the wrong one is the most expensive line you can write in a robots.txt file:
- Search bots fetch pages so the assistant can answer with current sources. OAI-SearchBot surfaces websites in ChatGPT's search features, Claude-SearchBot works on search result quality for Claude, and PerplexityBot surfaces and links websites in Perplexity results, which Perplexity states is not used to crawl content for foundation models. Block one of these and you disappear from that assistant's answers.
- User triggered fetchers visit a page because a person asked for it: ChatGPT-User, Claude-User and Perplexity-User. OpenAI writes that robots.txt rules may not apply to ChatGPT-User because the action is initiated by a user, Perplexity writes that Perplexity-User generally ignores those rules, and Anthropic states that its bots honor robots.txt directives.
- Training bots and tokens govern whether your content may be used to train models: GPTBot, which OpenAI documents as crawling content that may be used in training its foundation models, ClaudeBot, which Anthropic documents as collecting web content for its models, and the Google-Extended token.
robots.txt
The plain text file at the root of your site that tells crawlers what they may fetch. OpenAI, Anthropic and Perplexity all document it as the way to control their crawlers, so it is where the three families above become a set of deliberate decisions instead of an accident.
llms.txt
An optional plain text file, proposed at /llms.txt, that lists your pages for AI systems. Google states that you do not need to create new machine readable files, AI text files, markup or Markdown to appear in Google Search, including its generative AI capabilities, because Google Search does not use them and ignores them (Google Search Central). Publish one if a tool you use asks for it, and fix retrievability and quotable content first.
A page can also fail for a reason none of these terms covers: it renders only in the browser, so the crawler receives an empty document. The mechanics of that failure, stage by stage, are in How AI models choose which brands to cite.
How AI visibility is measured
Citation
A source the model names or links in its answer. In AI search a citation is normally a link you can click and audit. In an answer written from training memory there may be no link at all, which is why citations alone undercount your presence.
Mention
Your brand named in an answer, with or without a link. The mention is the commercially relevant event: the reader takes the shortlist from the answer even when they never click anything.
Mention rate
The share of runs of a fixed set of prompts in which a model names your brand. It is the closest equivalent to a ranking in AI search, and it has to be a rate rather than a single observation. The method for producing one by hand, and the point at which it stops being practical, is in How to check if ChatGPT mentions your brand.
Share of voice
Your mentions as a proportion of all brand mentions across the same prompts and the same period. It is the comparative half of the picture: a mention rate tells you how often you are named, share of voice tells you how much of the conversation is yours rather than a competitor's.
Prompt set
The fixed list of realistic questions you track, written the way customers actually ask them and deliberately without your brand name in them. Keep the set stable over time: if you change the questions, you change the numbers, and you lose the trend.
Non-determinism
The property that makes AI visibility awkward to measure: the same prompt, sent to the same model on the same day, can return a different shortlist. Repeated sampling across models and over time is the only honest answer to it.
Position and prominence
Where your brand appears in an answer and how much of the answer it carries. Being named first with a sentence of context is worth more than appearing last in a list of alternatives, so a serious measurement records both rather than a yes or a no.
Tracking all of this by hand across four models is where the practice usually stalls. Huntair SEO automates it on paid plans: you choose the prompts to track across ChatGPT, Gemini, Claude and Perplexity, and the mention rates and competitor comparison sit next to your keyword rankings on the Google SERP in one dashboard.
Content and entity terms
Answer-first content
A page or section that opens with a self contained answer of two or three sentences, then elaborates. It is the single most transferable GEO habit, because a short complete answer is the block a model can lift without risk.
Question shaped heading
A heading written as the question a reader would ask, such as "How much does it cost?" instead of "Pricing information". It helps a model decide which section of your page answers the prompt in front of it.
Entity
What a model believes your brand is: one name, one category, one line of description, consistent everywhere it looks. An ambiguous entity gets hedged or skipped when a model wants to sound confident, so consistency across your site, your profiles and the press behaves like a ranking factor.
Structured data
Schema.org markup that describes your pages in machine readable form, for example Organization, Article and FAQPage. Google states that structured data is not required for generative AI search and that there is no special schema to add for it (Google Search Central), while it stays useful for the rest of your SEO.
FAQ
Is AEO the same thing as GEO?
In practice, almost. AEO came first and emphasizes giving a machine a clean, liftable answer; GEO emphasizes being selected and cited by a generative model. The work they describe overlaps so heavily that running two separate programs makes no sense. Pick the term your team understands, define it in one sentence, and measure it.
Which term should I use with my team?
Use GEO when the goal is being cited inside AI answers, and keep SEO for ranking in search results. The two names then map onto two different metrics, a mention rate and a ranking, which is the only distinction that changes what you do on Monday morning.
Do I need a different strategy for each AI model?
The content work is shared: readable pages, answer-first sections, a consistent entity. The technical work is per provider, because each of them runs and documents its own search crawler, so a page one of them cannot fetch cannot be cited by that assistant. Measurement is per model too, since visibility on one does not transfer to the others.
Where should I start if all of this is new?
Read What is GEO? A practical guide for the concept, then GEO and SEO compared for what carries over from work you have already done. How AI models choose which brands to cite explains the mechanism, and How to check if ChatGPT mentions your brand gives you a first measurement to work from.