Reference · 12 min read
The GEO glossary: 32 terms, defined once
Plain definitions of the terms you meet in AI visibility work: GEO, AEO, answer engines, citations, mentions, share of answer, grounding, and the crawlers behind them.
· By the Askwords team
Three groups invented the vocabulary of AI visibility at the same time: search marketers, the AI companies, and the tools in between. So some things have four names and some have none. This page is our attempt to pin down what each term means, in a couple of sentences, with a link where we've written more.
Every entry has its own anchor, so you can link straight to one definition (say, share of answer) from a report or a brief. We grouped terms by where you'll run into them rather than A to Z. The contents list on the left jumps between groups.
If you can't find a term, it's probably a synonym of one that is here. GEO, AEO and LLMO all describe the same work. And whatever the tool calls it, there are really only three results: cited, mentioned, absent.
The field
- AI visibility
- Whether AI answer engines name your business when a customer asks them a buying question, and how favourably. You measure it per question and per engine, because the same brand can be recommended by ChatGPT and missing from Google's AI Overviews for the exact same query. SEO vs. GEO
- GEOGenerative engine optimization
- The work of getting your business cited and recommended inside AI answers. It builds on SEO, since the pages still have to be crawlable and believable, but the goal shifts from a ranking on a results page to a mention in the one answer the customer actually reads. What is GEO?
- AEOAnswer engine optimization
- An older name for more or less the same thing as GEO, from when the answer engines in question were featured snippets and voice assistants. Today the two overlap almost completely. Some people use AEO for the question-and-answer formatting part and GEO for the whole practice.
- LLMOLarge language model optimization · LLM SEO
- Yet another label for getting large language models to retrieve and cite your content. Mostly used by agencies and tool vendors. If an article uses GEO, AEO and LLMO in the same breath, it's describing one job.
- SEOSearch engine optimization
- The established practice of earning a spot on a search results page. Most of it still matters for AI answers: an engine can't cite a page it never crawled, and it leans toward sources that already look authoritative. What changes is how you keep score. SEO vs. GEO
The answer engines
- AI answer engineAnswer engine
- Any system that replies to a question with a written answer rather than a list of links: ChatGPT, Google's AI Overviews and AI Mode, Gemini, Perplexity, Claude, Copilot. Each finds its sources differently and has its own habits about which ones it names, which is why you have to check them one by one. Five answer engines, five shortlists
- AI OverviewsAIO · formerly SGE
- The AI-written summary Google puts above the normal results for a lot of queries. It cites pages from Google's regular index, so it has the biggest audience of any AI answer and the closest tie to the search presence you already have.
- AI Mode
- Google's conversational search tab: the whole page is an AI answer and follow-up questions keep the context. It uses the same index as Search but fires several sub-queries per question (see query fan-out), so its shortlist can differ from the AI Overview for the same words.
- ChatGPT search
- ChatGPT looking things up on the live web before it answers, with citations. It retrieves through OpenAI's own crawler (OAI-SearchBot) and partly through Bing's index, which is why Bing Webmaster Tools matters more for ChatGPT visibility than most people expect.
- Perplexity
- An answer engine built around citations, where every claim links to a source. Smaller audience than Google or ChatGPT, but a research-minded one, and the easiest engine to see what it cited and why.
- Gemini
- Google's assistant, separate from Search. It answers from what the model knows plus Google grounding. If Google already knows your business well (a Business Profile, a Knowledge Panel), this is the engine where that shows most.
How answers get made
- RetrievalRAG · Retrieval-augmented generation
- The step where an answer engine fetches fresh pages before it writes, instead of relying on what the model memorised in training. Nearly every citation you can earn comes through retrieval. That's why crawler access and up-to-date pages matter more than trying to get into a model's training data. Where an AI answer gets its sources
- Grounding
- Tying an answer to specific retrieved sources, so the model says what those pages say instead of guessing. A grounded answer cites. An ungrounded one names brands from memory, and memory is exactly where small businesses tend to be missing.
- Query fan-out
- Splitting one question into several searches and combining the results. Google's AI Mode does this openly: "best dentist Dublin" becomes separate queries about reviews, prices, opening hours and neighbourhoods. So the page that wins is often the one that answers a sub-question properly, rather than the one targeting the original phrase.
- CitationCited
- A link or explicit source attribution inside an AI answer that points at your page. It's the best outcome a page can get: you're named, and the reader can click through. Askwords records a brand as cited when the answer names it and links to its site. Close a citation gap
- MentionMentioned
- Your brand's name is in the answer, but without a link, or without being the recommendation. One of twelve names in a list is a mention. Worth tracking on its own, because a mention often turns into a citation once your pages give the engine something worth linking.
- Absent
- The answer to a question in your market doesn't contain your brand at all, though it may well contain your competitors. It's the most common result for a small business on its first check, and also the one with the clearest fix list. Absent: what to fix first
- Recommendation
- The engine doesn't just mention you, it points the reader at you: you're the answer, or on the short list it endorses for that question. This is what the whole practice is for. Usually it takes being cited by sources the engine already trusts, not only by your own site.
- Hallucination
- A confident, wrong statement in an AI answer: a price you don't charge, a location you closed, a service you never offered. Hallucinations about your business are a visibility problem in their own right. The fix is the same as for being absent: publish clear, crawlable facts so the engine retrieves instead of guessing.
- Non-determinism
- Ask an answer engine the same question twice and you may get two different shortlists. One run is an anecdote. A change only means something if the same question, asked again, still shows it. That's why a serious tracker re-asks before it reports a move. Read a notification
Measuring it
- Buyer questionPrompt · Tracked prompt · Askword
- A question a real customer would ask an AI while deciding what to buy, in their words and for their market: "best family dentist in Dublin 6?" rather than "dentist". Tracking tools differ mostly in how they pick these. A made-up prompt nobody types gives you a number that measures nothing. Run a real audit
- AI visibility score
- One number that rolls up cited, mentioned and absent results across questions and engines, usually weighted by how many people use each engine. Handy for trends and comparisons, risky on its own. Always look at the per-question results behind it.
- Baseline
- Your first complete, dated measurement of where you stand: which questions, which engines, which result for each. Every later claim that things improved is a comparison with this, so write it down. Memory doesn't count.
- RecheckVerified change
- Asking the same question again before reporting that a result changed, to tell a real move from the model having a different afternoon. A notification that survived a recheck is a signal. One that didn't is noise.
- AI referral traffic
- Visits that land on your site from an AI answer, recognisable by the referrer (chatgpt.com, perplexity.ai, gemini.google.com and so on). It undercounts, because plenty of answers get read and never clicked, but it's the most direct proof you have that citations turn into customers.
Under the hood
- AI crawlerAI bot · User agent
- The automated fetcher an AI company sends out to read web pages. Each one identifies itself with a user-agent token (OAI-SearchBot, PerplexityBot, ClaudeBot) and most obey robots.txt. The distinction that matters is between crawlers that feed answers and crawlers that feed training. Block the first kind and you lose citations. Block the second and you don't. The AI crawler list
- robots.txt
- The plain text file at the root of your site that tells crawlers which paths they may fetch. One over-broad rule ("block all AI bots") is the most common way a business vanishes from AI answers without ever meaning to. Can AI crawlers reach your site?
- llms.txt
- A proposed convention: a Markdown file at /llms.txt that summarises your site for AI assistants. No major answer engine has confirmed it reads one, so it's cheap to ship and unwise to count on. We ship one. We don't expect much from it.
- Entity
- Something the engine recognises as a distinct, named thing (your company, your product) rather than a string of words. Weak entity signals, meaning no consistent name, address, description and profiles across the web, are a common reason a brand is absent even when its pages are fine.
- Structured dataSchema.org · JSON-LD
- Machine-readable facts embedded in a page: this is an Organization with this name, this address, these profiles; this page is an Article published on this date. It won't make an engine cite you, but it removes any doubt about what you are, which is what an entity needs.
- Third-party proofCorroboration
- Independent pages that say the same things your site says: reviews, directories, press, industry lists. Answer engines trust sources that agree with each other, so a claim that lives only on your own site is one the engine hedges on or skips.
Cited, mentioned or absent. The rest is context.
Most of this page exists to explain those three words. The free check gives you them for your own business, per question and per answer engine, in about a minute.
Check three buyer questions free Keep reading
Sources and further reading
- Google Search Central: AI features and your website
- OpenAI: GPTBot, OAI-SearchBot and ChatGPT-User
- Schema.org: DefinedTerm
Askwords measures a selected set of buyer questions against fresh answers from each engine. AI answers vary by model, time, context and personalization, so a check is evidence of the answers observed, not a universal judgement about a brand.