Clear language for the new discovery layer.

An AI system that replaces lists of links with generative answers.

The degree to which a brand is mentioned and cited in AI answers.

Crawlers that fetch web content, e.g. GPTBot, PerplexityBot, ClaudeBot. Allowing them is a prerequisite for being cited.

AEO (Answer Engine Optimization) is optimizing content so that answer engines can extract a correct, self-contained answer from it and attribute that answer to you. It overlaps heavily with GEO; where the two are distinguished, AEO emphasises the extractable answer block and GEO emphasises whether the brand gets named and recommended.

AI Overviews are the generated summaries Google places above traditional results for some queries, with links to supporting sources. They sit inside the search results page rather than in a chat interface, which makes them the answer surface most likely to affect existing search traffic.

AI referral traffic is the visits that arrive from an assistant's interface, identifiable in analytics by referrer hosts such as chatgpt.com or perplexity.ai. It is a fraction of the exposure, because most answers are read without a click.

Answer form is the shape an assistant chooses for its reply: a ranked list of named options, an unranked list, or guidance that names nobody at all. It is a property of the question and the engine, not of any brand.

A buyer question is a complete, natural question a real buyer would type into an assistant, such as which option suits a stated situation. It is the unit of measurement in GEO, and it is not a keyword: keywords are fragments, buyer questions are sentences with a situation in them.

Buyer intent is what the person asking is trying to decide: understand a category, compare options, judge whether something is worth it, verify trust, or choose what to buy now. In GEO it is attached to each tracked question so results can be read by decision stage.

Brand normalization is resolving the different ways a brand is written in answers to a single entity: casing, punctuation, legal suffixes, a product line named instead of the company. Without it the same brand appears as several rows in a share-of-voice table.

The share of citation sources in relevant AI answers that belong to your pages/domain.

A citation is a source reference inside an AI answer that resolves to a specific URL. It is not the same as a retrieval candidate: an assistant typically retrieves a dozen or more pages and cites only a few of them in the text it writes.

A cited domain is a website that an AI answer references as a source. Grouping citations by domain and by domain type (community, directory, publisher, brand site, research) shows where an answer engine is getting its picture of a category.

ClaudeBot is Anthropic's crawler for collecting web content that may contribute to training its models. It is one of three separately named agents: Claude-SearchBot crawls to improve the search results shown to Claude users, and Claude-User fetches a page because a person's question requires it.

Content freshness is how recently a page was published or substantively updated, and whether that date is stated in a form a machine can read. Retrieval systems use recency signals when a question implies a current answer.

A competitor set is the list of brands you consider alternatives to yours. In AI measurement it should be an input for reporting, never the filter that decides which brands get counted in an answer.

Extractability is how readily a machine can lift a correct, self-contained statement out of a page. It depends on the page being served as real HTML, on headings that match the questions being answered, and on paragraphs that stand alone.

An entity is a thing a machine treats as one identifiable subject: a company, a product, a person. A brand becomes an entity when many independent sources describe the same name, domain and category consistently enough that a model can resolve them to one subject.

E-E-A-T is Google's framing of the qualities its raters look for in content: first-hand experience, expertise, authoritativeness and trust. It is a quality guideline rather than a computed score, and it long predates answer engines.

FAQ schema is FAQPage structured data that marks a set of questions and their answers on a page. It gives an answer engine a pre-segmented, question-shaped block instead of asking it to infer where an answer begins and ends.

Getting a brand mentioned, cited and recommended in AI generative answers.

GPTBot is OpenAI's crawler for gathering web content. It sits alongside agents with different jobs, including OAI-SearchBot for search indexing and ChatGPT-User for fetches triggered by a user's request, and other vendors run their own set such as PerplexityBot, ClaudeBot and Google-Extended.

Google-Extended is not a crawler. It is a usage control: it determines whether content Google has already crawled may be used to train Gemini models and to ground answers in Gemini Apps and Vertex AI. Google documents that it does not affect inclusion or ranking in Google Search.

Grounding is connecting a model's answer to retrieved documents so that its claims trace to sources rather than to memory alone. When grounding is on, the answer usually carries citations; when it is off, the model answers from what it absorbed in training.

A hallucination is a confident statement a model makes that is not supported by any source. For brands the common form is not invention from nothing but a plausible wrong detail: an outdated price, a feature that belongs to a competitor, a market you left.

A knowledge graph is a structured store of entities and the relationships between them. Public graphs such as Wikidata are read by many systems, which is why an entry there can influence how a brand is identified far beyond the site that hosts it.

A root-level content index file aimed at large models (emerging standard).

LLMO (Large Language Model Optimization) is an umbrella term for influencing how language models represent a brand, covering both what a model absorbed during training and what it retrieves at answer time. In practice most published LLMO advice is about the retrieval half, because that is the half anyone can affect this quarter.

Mention rate is the share of measured answers in which a brand is named. The denominator is answers that completed, not questions asked: a model that failed to respond produces no evidence either way and must be left out.

A measurement methodology is the fixed set of choices behind a visibility number: which engines, which question set, whether web search is on, and what counts as a mention. Changing any of them changes the ruler, not the thing being measured.

OAI-SearchBot is the OpenAI agent that crawls for search indexing, as distinct from GPTBot, which collects content that may be used for model training, and ChatGPT-User, which fetches a page in response to a person's request.

PerplexityBot indexes pages so they can be surfaced and linked in Perplexity answers. Perplexity's documentation states it is not used to crawl content for foundation models. A second agent, Perplexity-User, visits a page because a user's question requires it.

A prompt set is the fixed list of buyer questions a brand tracks over time. It is the denominator behind every visibility percentage, which is why changing it silently invalidates the trend it appears on.

robots.txt is a file at the root of a site that declares which crawlers may fetch which paths. It is a declaration that well-behaved crawlers honour, not an enforcement mechanism, and it says nothing about what happens once a page has been fetched.

RAG is the pattern of retrieving documents relevant to a question and giving them to a model as context before it writes an answer. Most answer engines that show sources are doing some form of this.

Share of voice is a brand's portion of all brand mentions across a fixed set of answers. It only holds if every brand the answers named is counted the same way, including brands that were never on your competitor list.

Schema markup is machine-readable data embedded in a page, usually as JSON-LD, that states what the page is about in a fixed vocabulary: an organisation, a product, a price, an author, a date. It does not change what a human sees.

A self-referential question is one that contains the brand's own name, such as asking whether a named company is legitimate. An assistant will of course mention a brand you just named, so these questions cannot measure whether it surfaces on its own.

Sentiment is how favourably an answer describes a brand when it names it: recommended without reservation, mentioned with a caveat, or named as the option to avoid. It is separate from whether the brand was named at all.

Source diversity is how many distinct domains, and how many kinds of domain, an answer engine draws on for a category. It is measured by counting distinct questions a domain was cited for, not by raw citation volume.

Unmeasured means no answer was collected for a question and model pair; zero means an answer was collected and the brand was absent from it. Rendering the first as the second turns missing evidence into negative evidence.

A zero-click result is one where the user gets what they needed without visiting any source. In an answer-engine context this is the normal case rather than the exception, since the answer is the product.