The AI search glossary.

Practices that make content more likely to be used as a direct answer — originally featured snippets and voice assistants, now also AI assistants. Overlaps heavily with GEO.

• GEO

A bot that fetches web pages for an AI company, either to train models or to retrieve pages for live answers. Examples include OpenAI’s GPTBot and OAI-SearchBot, Anthropic’s ClaudeBot and PerplexityBot. Most can be allowed or blocked in robots.txt.

• robots.txt

A conversational search experience in Google Search that answers questions with AI-generated responses and links, and supports follow-up questions. Distinct from AI Overviews, which appear on regular results pages.

• AI Overviews

AI-generated summaries that Google shows at the top of some search results, with links to the sources used. Formerly tested as the Search Generative Experience.

Visits to your website from links inside AI assistant answers, usually identified by the referring domain (such as chatgpt.com or perplexity.ai) or by UTM parameters the assistant adds.

• AI referral traffic

How often, and how prominently, AI assistants mention or recommend a brand when people ask questions in its category.

• Methodology
• Answer monitoring

Another name that should count as your brand — a product name, abbreviation, former name or common misspelling. In Mentioned, aliases are matched when reading answers.

• How answers are read

A system that responds to a question with a direct, composed answer rather than a list of links. ChatGPT, Perplexity, Claude, Gemini and Google’s AI features all act as answer engines.

The mean rank of a brand across answers where it appears in a list of options. Answers that mention the brand outside a list, or not at all, don’t affect it.

• Definitions

A question a potential customer would ask an AI assistant while choosing a product, such as “best CRM for a small sales team”. Tracking a set of buyer prompts over time is the basis of AI visibility measurement.

• Buyer prompts
• Prompt intent

A link an AI answer attributes information to. Citations show which pages an engine relied on, and are one of the clearest signals of which sites shape answers in a category.

• Citation analysis

Finding the vendors AI assistants recommend in your category that you don’t yet track — often different from the competitors you see in traditional search.

• Competitor tracking

A range around a measured value that likely contains the true value. Mentioned reports a 95% margin of error next to each visibility score, so changes smaller than the range can be treated as noise.

• Confidence ranges

The maximum amount of text, measured in tokens, that a language model can consider at once — including the question, any retrieved sources and the answer it writes.

• Token

Experience, Expertise, Authoritativeness and Trustworthiness — concepts from Google’s Search Quality Rater Guidelines used to describe what makes content reliable. Not a direct ranking score.

A distinct, identifiable thing — a company, product, person or place — that search engines and language models can recognise and connect to facts. Clear, consistent entity information helps models describe a brand correctly.

• Knowledge graph

A Google search result that quotes a short answer from a web page above the regular results. An early form of answer-first search.

A website that AI answers cite when recommending your competitors but never when mentioning you. Gap domains are often the most practical places to earn a listing, review or mention.

• Citation analysis

Improving how a brand and its content are represented in answers from generative AI search engines. The term was popularised by a 2023 academic paper of the same name.

• AEO
• LLMO

Tying a model’s answer to retrieved sources — such as live web search results — so that statements can be supported and cited, rather than relying only on what the model learned in training.

• RAG

When a language model produces plausible-sounding information that is false or unsupported — for example, a product feature that doesn’t exist or a source that doesn’t say what is claimed.

The point after which a model’s training data contains no information. Web search and grounding let answers include newer information.

A structured database of entities and the relationships between them, such as Google’s Knowledge Graph or Wikidata. Search engines use knowledge graphs to understand what a brand is and how it relates to others.

• Entity

A machine-learning model trained on large amounts of text to predict and generate language. LLMs power AI assistants such as ChatGPT, Claude and Gemini.

A term used much like GEO: improving how often and how accurately large language models mention a brand or use its content.

• GEO

A proposed convention, introduced in 2024, for a Markdown file at /llms.txt that gives language models a concise guide to a site’s most useful content. Support among AI providers varies.

• Our llms.txt

In AI visibility tracking, an answer that names a brand as an option, recommendation or subject. A passing reference usually doesn’t count.

• What counts as a mention

The share of answers that mention a brand, out of all answers measured. The unweighted counterpart of a visibility score.

• Definitions

A change to the model behind an AI product or API. Updates can shift many answers at once, which is why sudden moves on a single engine deserve a closer look before acting.

Tailoring answers to a user through memory, chat history, account settings or location. It’s one reason a consumer app can answer differently from the same model’s API.

• Limitations

The text given to a language model. In AI visibility tracking, a prompt is a question asked to each engine on a schedule.

• Buyer prompt

The purpose behind a buyer prompt — for example “best X for Y”, “X vs Y”, “alternatives to X” or “how to …”. Tracking a mix of intents shows where in the buying journey a brand appears.

• Buyer prompts

A technique where an AI search system runs several related searches for a single question and combines the results into one answer.

An approach where a system first retrieves relevant documents — from the web or a private index — and gives them to a language model to write its answer. Web-search features in AI assistants work this way.

• Grounding

A text file at the root of a website that tells crawlers which paths they may fetch. Many AI crawlers honour it, and some providers offer separate tokens for training versus search.

• AI crawler

Checking a different slice of prompts on each run so that every prompt is still covered within a set window. In Mentioned, Enterprise daily tracking uses rotation alongside pinned priority prompts.

• Schedules

Asking the same prompt more than once to measure how much answers vary. More samples narrow the margin of error at a higher cost.

• Confidence interval

Machine-readable descriptions of a page’s content using the schema.org vocabulary, usually embedded as JSON-LD — for example Organization, Product or FAQPage. It helps search systems understand entities and facts.

The tone an answer takes toward a brand. Mentioned scores it from −1 (very negative) to +1 (very positive) for each answer that mentions you.

• Definitions

How often a brand appears relative to its competitors. In AI visibility, it’s typically the number of answers that mention each brand across the same set of prompts.

• Competitor tracking

A page an AI engine retrieves or cites while answering. Aggregated by domain, sources show which sites shape what AI says about a category.

• Citation

A setting that controls how random a model’s output is. Higher values give more varied answers; lower values make output more consistent. Consumer apps don’t expose it.

The unit of text a language model reads and writes — often a word or part of a word. Model limits and API prices are usually expressed in tokens.

Mentioned’s 0–100 measure of AI search visibility: the share of answers that mention a brand, weighted by its position in any list, over a rolling 7-day window.

• The formula

An API capability that lets a model search the web while answering and return the sources it used. Each major AI provider offers one in some form.

• Engines

A search that ends without a click to another website because the answer was shown directly — in a snippet, an AI summary or an assistant’s reply.