Glossary
The vocabulary of AI answers
40 terms used across the docs, the API reference and the blog, covering both sides of this market: the measurement vocabulary buyers report in, and the API vocabulary engineers build with. Each definition links to the page that goes deeper.
- AEO (answer engine optimization)
- The practice of getting content quoted, cited and recommended by answer engines rather than ranked by them. AEO and GEO name the same work, measured with mention rate and citation share instead of keyword positions. What AEO is
- AI crawler
- A bot that fetches web pages for an AI product, either to build an index (GPTBot, ClaudeBot, OAI-SearchBot) or to read a page live while answering a user (ChatGPT-User, Perplexity-User). Which ones a site allows in robots.txt decides which answers can quote it. AI crawlers and robots.txt
- AI Mode
- Google's conversational search surface, where a question is answered in prose with follow-ups, product blocks and place blocks instead of a page of links. It is a separate surface from AI Overview and cites differently. AI Mode API
- AI Overview (AIO)
- Google's AI-generated answer block shown above organic results for some searches, with links to the pages it drew on. Whether it appears depends on the keyword, the market and the device. AI Overview fields
- AI visibility
- How often and how favorably AI answers name, recommend and cite a brand across a set of monitored prompts. It is measured over a prompt set and a date range, not on a single answer. AI visibility tracking
- Answer engine
- A product that replies to a question with a written answer instead of a list of links: ChatGPT, Perplexity, Gemini, Copilot, Grok, Google's AI Mode and AI Overview. Also called generative engines. Engines covered
- Answer rank
- The place a brand takes in an answer's recommendation order, named first rather than fourth. On engines that return entities it is read from their order; elsewhere it is read from where the brand appears in the text. Rank tracking in AI answers
- API key
- The secret sent with every request to identify the account and bill it. One key grants every endpoint and keys have no scopes, so a leaked key is revoked and replaced rather than narrowed. Authentication and errors
- Async task
- A request that is queued instead of answered on the open connection: the API returns a task id immediately, and the result arrives by webhook or is collected later. Async is the cheaper and higher-volume mode. Async tasks and batches
- Canonical URL
- The address a site declares as the preferred one for a page, through a rel=canonical link or a redirect. When several addresses serve the same content, citations split across them unless one is canonical. How engines choose citations
- Citation
- A source an engine links as evidence for part of its answer, shown as a numbered or bracketed reference. A citation means the page was retrieved and used, which is a stronger signal than being named. Sources and citations
- Concurrency
- The number of requests an account may have in flight at the same time, set by its plan. Synchronous calls and running async tasks draw on the same pool; tasks beyond it wait in the queue. Rate limits and concurrency
- Credit
- The unit requests are priced in. Each request costs a fixed number of credits set by the engine and the options enabled, and only successful results are charged. Credits and billing
- Entity
- A named thing in an answer, a brand, product or organization, returned as a typed record rather than left in the prose. Mention tracking reads entities where an engine provides them. Entities in ChatGPT answers
- GEO (generative engine optimization)
- The practice of improving how often and how favorably AI answers mention and cite a brand. Success is measured as mention rate and citation share across a prompt set, not as a position in a ranking. What GEO is
- Grounding
- The retrieval step where an engine searches an index, reads the pages it finds and writes its answer from them. The index behind that step decides which pages can be cited at all. Grounding by engine
- Idempotency key
- A client-chosen identifier attached to a request so that retrying it returns the original result instead of doing the work twice. It is what makes a timed-out submission safe to repeat. Idempotency keys
- Knowledge graph
- The entity panel Google shows beside results for a person, place, company or product, with attributes drawn from its entity database. It is the clearest read on what a search engine believes about a brand. Knowledge graph fields
- llms.txt
- A proposed Markdown file at a site's /llms.txt path that summarizes the site and links its most useful pages for language models. It is a proposal, not a web standard, and no major AI search vendor documents reading it from your site. What llms.txt does
- Local pack
- The map and three-business block Google shows for searches with local intent, ranked separately from organic results. Its membership changes with the searcher's location, so it has to be queried per market. Local pack fields
- MCP (Model Context Protocol)
- An open protocol that lets an AI agent call external tools through a standard interface. An MCP server exposes this API's engines as tools an agent can invoke while it reasons. MCP server
- Mention
- A brand or product named in an answer's text, whether or not the answer cites it as a source. A mention without a citation means the model learned the brand from someone else's page. Mentions versus citations
- Mention rate
- The share of prompts in a monitored set whose answers name a given brand, tracked per engine and market. It is the headline AI-visibility metric because it is comparable across competitors. AI visibility tracking
- Organic result
- A ranked link on a search results page that was not paid for, with its title, URL, snippet and position. Position continues across pages, so result eleven is the first result of page two. Organic result fields
- People Also Ask (PAA)
- The expandable related questions Google interleaves with results, each with a short extracted answer and a source. The question set is a direct read on what searchers ask around a topic. People Also Ask fields
- Prompt set
- The fixed list of natural-language questions monitored on a schedule, the answer-engine equivalent of a keyword list. Prompts are usually grouped by intent: category, comparison, problem and branded. Designing a prompt set
- Query fan-out
- The several reworded web searches an engine runs behind the scenes before it answers one prompt. The pages that rank for those searches form the pool the answer cites from. Query fan-out explained
- RAG (retrieval-augmented generation)
- The pattern of retrieving documents at answer time and writing the answer from them, rather than from model weights alone. Consumer answer engines are RAG systems whose corpus is the web. How engines choose citations
- Rate limit
- The cap on how many requests an account may send per time window, reported back in response headers. Exceeding it is rejected with a retryable error rather than queued. Rate limits and concurrency
- Sampling
- Measuring a prompt set by running part of it, or running it on some days, and inferring the rate for the whole. Sampling trades precision for cost, and the margin of error follows the number of observations. Sampling versus census
- SERP
- The search engine results page: everything returned for one query, including organic links, ads, packs, panels and AI blocks. A SERP is specific to the query, the market and the moment it was requested. What a SERP contains
- SERP feature
- Any block on a results page that is not a plain organic link: ads, People Also Ask, the local pack, the knowledge graph, video carousels, AI Overview. Which features appear is itself a ranking signal worth tracking. Tracking SERP feature changes
- Shopping card
- A product block inside an answer or results page carrying a title, price, merchant, rating and link. For retail brands it is the shelf: which products an engine puts in front of a buyer, and at what price. Shopping results
- Source position
- The order a cited page takes in an answer's source list, the number a reader sees next to it. Position one is the lead citation and carries most of the attention. Perplexity sources
- Structured data
- Machine-readable markup on a page, usually schema.org in JSON-LD, describing what the page is about. It helps search engines classify a page; it does not by itself make an AI answer cite it. Structured data and AI visibility
- Sync versus async
- A synchronous call holds the connection open and returns the answer in the response, for a surcharge. An async task queues the work and delivers the result by webhook at the base price, which is the right shape for volume. Choosing sync or async
- Volatility
- How much an answer or ranking changes between runs of the same prompt. High volatility means a single run tells you little, and a metric needs repeats before a movement counts as a change. Answer volatility
- Webhook
- An HTTP callback the API sends to a URL you own when a task finishes, carrying the result and a signature to verify it. Webhooks replace polling and must be handled idempotently, because a delivery can repeat. Webhooks