Glossary

AI visibility glossary

41 terms of AI visibility and GEO, defined plainly. The words we use in our reports mean exactly what is written here.

Core concepts

GEO (generative engine optimization): The work of getting a brand named, and its pages cited, in answers generated by AI engines on the questions its buyers ask. The term comes from a 2023 research paper presented at KDD 2024. What is GEO? →
AEO (answer engine optimization): An older, overlapping term for optimizing content to be used as a direct answer, first for featured snippets and voice assistants, now also for AI engines.
AI visibility: How often, and how, a brand appears in AI-generated answers to a defined set of questions: named, cited as a source, or absent.
AI engine: A product that writes an answer to a question instead of (or on top of) returning a list of links: ChatGPT, Claude, Gemini, Perplexity, Google AI Overviews and AI Mode. Also called an answer engine.
LLM (large language model): The model that generates the text of an AI answer. AI search products combine an LLM with live web retrieval.
Google AI Overviews: The AI-generated summary Google shows at the top of some search results, with links to supporting pages.
Google AI Mode: Google’s conversational search experience, separate from regular results. Controlled by the same Search directives as AI Overviews.

Outcomes in an answer

Named: The brand appears in the text of the answer. Some tools call this a mention.
Cited (source citation): One of the brand’s pages is listed as a source of the answer. A brand can be cited without being named.
Source only: The brand’s page is cited but the brand is not named in the text. Counted separately from “named”, never added to it.
Absent: The brand is neither named nor cited in the answer.
Shortlist: The handful of vendors an AI answer names for a buying question. Being on it is the practical goal of GEO.

Measurement

Buying question: A question a prospect asks before contacting a vendor: “best tool for…”, “alternative to…”, “X vs Y”, “does X integrate with…”. The unit of work in GEO, where SEO uses the keyword.
Branded question: A question that already contains the brand’s name. Kept out of visibility totals because it inflates them.
Intent family: A group of buying questions with the same purpose (discovery, comparison, integration, compliance…). Results are read by family to see where a brand drops out.
Run: One execution of a question on one engine. Because answers vary, each question is run several times; ZivRank uses 5 runs per question and per engine.
Named rate: Answers naming the brand divided by all answers in the protocol, reported with a margin of error.
Cited rate: Answers citing one of the brand’s pages divided by all answers in the protocol.
AI share of voice: The brand’s mentions as a share of mentions of the brand and a validated list of competitors, on the same questions. Meaningless without the competitor list and the weighting. GEO KPIs →
Margin of error (confidence interval): The range within which a measured rate is likely to fall. A rate published without one cannot show whether a change is real.
Noise floor: The variation observed when nothing has changed, measured by running the same protocol twice. A change smaller than the noise floor is not a result. Why one answer is not a measurement →
Baseline measurement: The locked starting measurement of a program. Every later measurement replays exactly the same questions, engines and runs so results are comparable.
Re-measurement: A measurement that replays the baseline protocol identically after changes have been made.
Raw answers: The timestamped answers collected during a measurement. Handing them over lets anyone check the numbers in a report.
Pickup time: How long AI engines take to pick up a newly published asset. Observed at 3 to 8 weeks in our programs; re-measuring earlier understates the effect.

How AI engines work

Retrieval (RAG): Retrieval-augmented generation: the engine searches for pages at answer time and uses them to write and support its answer. This is how AI search products cite sources.
Grounding: Tying a generated answer to retrieved sources so it can be checked. Grounded answers show citations.
Query fan-out: A technique Google describes for AI Overviews and AI Mode: issuing several related searches across subtopics to find supporting pages for one question.
Training data: The text a model learned from before its cutoff date. It shapes which brands a model knows and how it describes them, even without live retrieval.
Hallucination: A confident but false statement in an AI answer, for example a wrong price, feature or certification attributed to a brand.
Zero-click: A search where the user gets what they need from the results page or AI summary without visiting any website.

Crawlers and site access

OAI-SearchBot: OpenAI’s crawler for ChatGPT search. Sites that block it are not shown in ChatGPT search answers, though they can still appear as navigational links.
GPTBot: OpenAI’s crawler for model training. Blocking it opts content out of training, not out of ChatGPT search.
PerplexityBot: Perplexity’s crawler for surfacing and linking sites in its answers. Perplexity states it is not used for model training.
Google-Extended: A robots.txt token that controls use of content for Gemini training and grounding. Google states it does not affect inclusion in Google Search.
llms.txt: A proposed plain-text file at the root of a site that summarizes its key pages for language models. A convention, not a standard that engines are known to require.

Content and sources

Answer-first content: Writing that puts the direct answer at the top of each section, followed by evidence. Easier for an engine to quote accurately.
Third-party source: A page not owned by the brand that describes it: comparison article, review platform, community thread, partner listing, industry publication.
Entity consistency: The same name, description, category and key facts about a company everywhere it appears. Inconsistency gives engines room to get facts wrong.
AI referral traffic: Visits that come from links in AI answers. Google Analytics 4 groups part of it in an AI Assistant channel; Google’s own AI features are reported in Search Console.
AI technical audit: A review of what prevents AI engines from reading and citing a site: crawler access, rendering without JavaScript, structure, structured data. Our technical audit →