GEO glossary
The words used when people talk about AI search, each in one or two sentences. Where a platform defines the term itself, we follow its documentation and link to it.
The basics
- Generative Engine Optimization
- GEO — Generative Engine Optimization — is the work that helps AI assistants such as ChatGPT, Claude, Gemini and Perplexity discover, understand, verify, cite and recommend a business.
- AI search
- Search in which an AI model reads web pages and writes the answer itself, instead of returning a list of links. ChatGPT search, Perplexity, Google AI Overviews and AI Mode work this way.
- AI assistant
- A chat product that answers in its own words — ChatGPT, Claude, Gemini, Perplexity, Grok, Copilot. When it searches the web before answering, it is also called an answer engine.
- Large language model
- The program behind an AI assistant, trained on large amounts of text to understand and write language. On its own it knows only what was in its training data.
- Training cut-off
- The date after which a model has learned nothing in training. Anything newer — a new service, a new address — reaches the answer only if the assistant searches the web.
- Google AI Overviews
- The AI-written summary Google shows above ordinary results for some searches, with links to the pages it drew on. Google shows it only when it judges it adds something to classic results.
- Google AI Mode
- A Google search mode for questions that need comparison or reasoning: the result is an AI answer with links to supporting sites, and the person can ask follow-up questions.
- Zero-click search
- A search that ends without a visit to any site, because the answer is already on the results page or in the chat. It is why being named in the answer matters, not only being linked.
How an AI answer is made
- Prompt
- The question or instruction a person types into an AI assistant. In our checks, prompts are written the way clients actually speak, not the way keyword tools list them.
- Query fan-out
- What AI search does with one prompt: it issues several related searches across subtopics and sources, then combines what it finds. A page can be found through a sub-question the person never typed.
- Retrieval-augmented generation
- The assistant first fetches documents — web pages, files — and then writes the answer from them. It is why current, readable pages matter more than what the model learned in training.
- Grounding
- Tying an AI answer to the specific sources it retrieved, so that each claim can be traced back. Grounded answers usually show links or citations.
- Hallucination
- A confident statement from an AI that no source supports — a wrong address, an invented service, a founder who does not exist. The same facts in many independent places make it less likely.
What we measure
- Mention
- Your company's name appearing in an AI answer, with or without a link.
- Citation
- A link or source reference in an AI answer that points to a specific page. A citation of someone else's page about you counts too.
- Recommendation
- An AI answer that names your company as an option for the client's need. It is the outcome GEO works towards; being mentioned is not the same as being recommended.
- AI Choice Visibility
- How often your company is named in AI answers to a fixed set of client questions, across engines and repeated runs. We report it as a share of answers, never folded into one composite score.
- Baseline
- The first measurement, taken before any work, with the questions, engines and dates locked. Every later result is compared against it.
- Known vs chosen
- The gap between an assistant describing your company correctly when asked by name, and naming it when asked for a recommendation without your name. The two are measured separately.
How AI learns who you are
- Entity
- Something a search engine or model recognises as one specific thing — a company, a person, a place, a product — rather than a string of words. GEO makes your company one clear entity.
- Knowledge graph
- A database of entities and the facts that connect them, such as Google's Knowledge Graph or Wikidata. Engines use it to check who is who.
- Structured data
- Machine-readable labels in a page, in the schema.org vocabulary, stating what it describes: an organisation, a person, a service, an article. It must match the visible text; no special AI markup exists.
- JSON-LD
- The format for structured data that Google recommends: a small block of code in the page listing the facts in a fixed structure. People do not see it; engines read it.
- sameAs
- A structured-data property listing your official profiles elsewhere — the business register, LinkedIn, Wikipedia — so engines can tell that they all describe the same company.
- Third-party sources
- Pages about you that you do not control: registers, directories, reviews, media, partners' sites. Engines weigh them heavily, because a company's own site is not independent evidence.
- AI crawlers
- The bots AI companies use to read the web, such as OAI-SearchBot and GPTBot (OpenAI), ClaudeBot (Anthropic) and PerplexityBot. Bots that fetch pages for answers and bots that collect training text can be allowed or blocked separately.
- robots.txt
- A text file at the root of a site that tells bots which pages they may fetch. A site that blocks OAI-SearchBot there is not eligible for ChatGPT search results.
- Google-Extended
- Not a separate bot but a robots.txt name that lets a site limit the use of its pages for AI training and grounding in some Google systems. Google Search and its AI features are controlled through Googlebot.
- llms.txt
- A proposed file that lists a site's key pages for language models. Google says no AI text files are needed for its AI features, and nothing shows that the file makes any assistant recommend a company; we treat it as an experiment.
See the terms on your own company
The free check shows mentions, citations and who is recommended instead — for your domain, on a known date.