Every term here comes with the calculation behind it, a labelled worked example, and the errors that most often make the metric misleading.
Agentic browsing is when an AI agent visits pages, clicks and reads on a user's behalf rather than answering from a search snippet. The agent, not the person, is the one reading your site — and what it can parse decides what the user is told.
AI Overviews are AI-generated summaries that Google renders inside the search results page for some queries, with a small set of source links attached. The defining property for measurement is that they are conditional: many searches produce no overview at all.
AI referral traffic is the sessions that arrive on a site from an AI assistant or answer surface. It is a real but partial signal: most AI answers are consumed without a click, so referrals count the minority of interactions that ended in a visit.
AI visibility is the degree to which a brand is mentioned, cited or recommended inside answers generated by AI systems such as ChatGPT, Perplexity, Gemini and Google AI Overviews. It is a sampled rate rather than a position, because the same question can produce different answers on two consecutive runs.
Answer engine optimization is the practice of structuring content so that an answer engine can extract, quote and attribute a direct answer from it. In everyday use it overlaps heavily with GEO; where people separate them, AEO is the page-level craft and GEO is the brand-level programme.
Answer-first content puts the direct answer to the page's question in its first two or three sentences, before context, story or setup. It is written so that the opening block can be lifted out and still be true, complete and attributable.
Brand sentiment in AI answers is how the brand is characterised in the answers that mention it: recommended, neutrally listed, or criticised. It is measured only across mentions, so it describes the quality of appearances rather than their frequency.
Chunk retrieval is the step where a system splits documents into passages and retrieves the individual passages that match a query. The unit an engine works with is therefore a block of your page, not the page as a whole.
The citation gap is the difference between how often a brand is mentioned in AI answers and how often its own domain is cited as the source. A wide gap means engines know the brand but prefer someone else's page as evidence.
Citation rate is the share of sampled answers that cite your own domain as a source. It separates being talked about from being used as the evidence behind the answer.
Competitive benchmarking in AI visibility is measuring named competitors on exactly the same prompt set, engines and sample size as your own brand. Because every rate depends on the instrument, comparisons are only valid when the instrument is identical.
Content freshness is how recently a page states it was created or updated, and how well that claim matches its actual content. Grounded answers favour sources that look current for questions whose correct answer changes over time.
An entity is a specific, identifiable thing — a company, a person, a product, a place — that a search or answer system can recognise and attach facts to. AI answers are assembled around entities, not keywords, which is why brand visibility work is entity work.
Entity disambiguation is the work of making a brand distinguishable from similarly named entities so that AI answers describe the right thing. It matters most for short, generic or near-homonym brand names, where a model can blend two products into one description.
Generative engine optimization is the practice of improving how often, how accurately and how favourably a brand appears inside answers generated by AI systems. It borrows research and content discipline from SEO, but it optimises for inclusion in one synthesised answer rather than for a position in a list of links.
Google AI Mode is a conversational search experience where the user asks follow-up questions and receives generated answers rather than a results page. It sits alongside AI Overviews, which appear above conventional results, and behaves more like a chat assistant than a SERP feature.
GPTBot is OpenAI's crawler, and it sits alongside a growing set of AI user agents — search-time fetchers, training crawlers and assistant browsers — each with a different purpose. Robots.txt rules decide which of them may read a site, and those rules have direct, verifiable consequences for visibility.
Grounding is the practice of generating an AI answer from documents retrieved at query time instead of from the model's trained-in memory. A grounded answer can cite what it used; an ungrounded one is reciting whatever the model absorbed during training.
A hallucination is an AI-generated statement presented confidently but unsupported by any retrieved source or by fact. In brand terms it usually appears as an invented feature, an outdated price, or a capability attributed to the wrong product.
A knowledge graph is a structured store of entities and the relationships between them: this company builds this product, this person founded this company, this product belongs to this category. Answer systems lean on graph-like knowledge to keep facts consistent across a generated answer.
llms.txt is a proposed convention: a markdown file at a site's root that lists its most important pages and summarises what they contain, intended to help language models find authoritative content quickly. It is a community proposal rather than a standard any major engine has committed to honouring.
A memory answer is one the model produces without retrieving anything, relying on what it absorbed during training. It reads exactly like a grounded answer and means something entirely different for measurement.
Mention position records where a brand appears relative to other named brands inside a single generated answer — first named, second, third. It is the closest analogue to a ranking in a category that has no rankings.
Mention rate is the share of valid, sampled AI answers in which a brand is mentioned at all. It is the base rate of AI visibility: before position, sentiment or citations mean anything, the brand has to appear.
no_surface is the status recorded when the AI surface being measured did not appear at all for a query — most often an AI Overview that Google chose not to generate. It is neither a mention nor a miss, and collapsing it into either one corrupts the metric.
Non-determinism means the same prompt can produce different answers on repeated runs, even with identical settings. It is a property of how these systems sample text, amplified by retrieval that returns a different source set minute to minute.
Prompt coverage is the share of tracked prompts in which a brand appeared at least once across the sampled answers. It measures breadth: how much of your question space you show up in at all.
Prompt intent classification is the step that sorts a tracked prompt into a class — brand, discovery, comparison, pricing, switching — before metrics are computed. It turns one flat average into segmented rates that describe different commercial problems.
A prompt set is the fixed list of questions a brand tracks across AI engines. It is the measurement instrument: every rate you report is a property of this list as much as of your content.
provider_error is the status for a sampled request that failed — a timeout, a rate limit, or an API error — so no answer was produced. It records a measurement failure, not a visibility outcome.
Retrieval-augmented generation is an architecture where a system retrieves relevant documents at query time and generates its answer from them. Every grounded AI answer you can influence with content is, in effect, a RAG pipeline pointed at the web.
Sample size, usually written N, is the number of times each prompt is run on each engine within a measurement cycle. Because AI answers are non-deterministic, N is what converts a series of individual answers into a rate with a usable margin of error.
Semantic similarity is the measure of how close two pieces of text are in meaning rather than in wording, usually computed from vector representations. Retrieval uses it to find passages that answer a question even when they share no keywords with it.
A SERP API is a service that fetches search engine result pages programmatically and returns them in structured form. It is how AI Overviews are tracked, since that surface lives inside the results page rather than behind a model API.
Share of search is a brand's share of branded search volume within its category — a demand measure built from what people type into a search engine. AI share of voice measures a different thing: how often a brand is named inside generated answers for category questions.
Share of voice in AI answers is the proportion of brand mentions among all tracked brands — yours plus the competitors you named — across the sampled answers. It answers "when this category comes up, how much of the airtime is ours?" rather than "how often are we mentioned?".
Source authority is the practical preference engines show for some pages over others when choosing what to cite. It is not a published score; it is an observable pattern in which sources keep appearing behind answers in a category.
Structured data is markup, usually JSON-LD using schema.org types, that states explicitly what a page is about: an article, a product, a question and its answer, an organisation, a defined term. It makes facts machine-readable instead of leaving them to be inferred from prose.
A zero-click search is one where the user gets what they needed from the results surface itself and never visits a website. AI answers push this further than featured snippets did, because a synthesised answer resolves multi-part questions in one place.
A zero-coverage prompt is a tracked question where the brand appeared in none of the sampled answers during a period. It is the practical work list that visibility measurement produces: a question your buyers ask that AI answers without you.
