60 AI search & automation glossary guides

Answer engine optimisation is the practice of making information easier for answer systems to find, understand and use when responding to a question. It covers clear answers, reliable evidence and accessible pages, not a guaranteed placement.

Agent2Agent is an open protocol for communication and collaboration between independent AI agent systems. It supports capability discovery and task exchanges without requiring agents to share their internal implementation.

Agentic commerce describes buying journeys in which AI agents help discover products, compare options or carry out authorised purchasing tasks. The degree of automation varies; payment, identity and consent still need explicit controls.

Agentic Commerce Protocol is an open specification developed by OpenAI and Stripe for interactions between AI agents and merchants during checkout. It standardises commerce-related exchanges; it is not a general permission for agents to spend money independently.

An AI agent is a software system that uses a model to choose actions and interact with tools toward a defined goal. Its access, instructions and approval boundaries determine what it can safely do.

AI citation rate is a defined measurement of how often a target source is linked in a sampled set of AI answers. Its meaning depends on the denominator, prompt set, engines and collection method; there is no universal formula.

AI evaluation is the systematic testing of model or agent behaviour against defined requirements. It can examine factual accuracy, retrieval quality, tool use, safety, cost and task completion using representative cases.

AI guardrails are controls that limit inputs, outputs or actions in an AI system. They can include access checks, spending limits, output validation and human approval; prompting alone is not a reliable security boundary.

AI Overviews are Google Search features that produce AI-generated summaries for some searches and may include links to supporting web pages. Availability and source selection depend on the query and Google’s systems.

AI referral traffic is website traffic attributed to visits from AI products or their linked answers. Some referrals lack identifiable source information, so analytics may undercount them or classify them as direct traffic.

AI search combines information retrieval with AI-generated answers. A system may search the web or a knowledge collection, select evidence and compose a response; not every model answer uses live search.

A backlink is a link from another website to a page on your website. Relevant editorial links can help discovery and contribute to search signals; a link is not proof of expertise by itself.

Bing Webmaster Tools is Microsoft’s service for website owners to review Bing-related site discovery, indexing and performance information. Its reports and submission tools do not guarantee rankings or AI citations.

A brand mention is a reference to a company or product in content or an AI answer. It may be positive, negative, inaccurate or unrelated; an unlinked mention is not a source citation or a qualified lead.

A canonical URL is the preferred address for a page when similar content is available at multiple URLs. A canonical tag expresses a preference to search engines; it is a signal, not an access restriction.

Chunking splits documents into smaller passages for retrieval or model processing. Useful boundaries preserve meaning and provenance; arbitrary cuts can separate an answer from its conditions or source.

An AI citation is a reference or link attached to a generated answer that attributes information to a source. A brand mention without a source link is not the same as a citation, and a citation does not prove the answer is accurate.

A context window is the amount of input and output information a model can consider in one interaction, measured in tokens. A larger window does not guarantee that every supplied detail is used correctly.

Conversion rate is the proportion of measured visits or interactions that complete a defined action. The denominator and action must be explicit; a booking-link click is not the same as a completed booking or a qualified enquiry.

Crawlability is the ability of a crawler to discover and fetch a website’s pages. Links, server responses, access controls and crawler rules affect it; being crawlable does not mean a page is indexed.

E-E-A-T stands for experience, expertise, authoritativeness and trustworthiness. Google uses these concepts in quality evaluation guidance; they are not a single score you can add to a page.

An embedding is a numerical representation of content used to compare patterns of meaning. Retrieval systems can use embeddings to find relevant passages even when wording differs; similarity does not establish factual correctness.

An entity is an identifiable person, organisation, place, product or concept. Consistent names, relationships and supporting information help readers and search systems distinguish one entity from another.

A featured snippet is a selected extract shown prominently in some Google search results to answer a query. Google chooses it from page content; it is distinct from an AI-generated summary.

Fine-tuning is additional training that adjusts a model using task-specific examples. It can improve recurring behaviours or formats; it is not the same as supplying current documents at answer time.

Generative engine optimisation is an industry term for improving the discoverability and usefulness of content in AI-generated responses. It overlaps with SEO and AEO; it is not a separate guaranteed ranking system.

Google AI Mode is a search experience that uses AI to support more complex queries and follow-up exploration, with links to relevant web sources. Its behaviour and availability can change independently of ordinary ranked results.

Google Search Console is Google’s service for verified website owners to inspect search performance, indexing and technical reports. It supplies evidence about Google Search, not a complete view of all AI engines or every lead.

Grounding connects a model’s response to supplied or retrieved evidence. It can improve factual reliability, but the evidence may still be incomplete, outdated or misinterpreted.

A hallucination is an AI-generated statement that is unsupported or incorrect but may sound plausible. It can include invented facts, sources or quotations, even when the answer is confidently written.

Human-in-the-loop design places a person at defined review or decision points in an AI workflow. The person can approve, correct or reject actions before they create material consequences.

Indexing is the process of analysing and storing a page in a search engine’s index after discovery. A page can be crawlable but not indexed, and an indexed page is not guaranteed to rank or be cited.

IndexNow is a protocol for notifying participating search engines when website URLs are added, updated or deleted. A notification helps discovery but does not guarantee indexing, and participation is engine-specific.

An internal link connects one page on a website to another page on the same website. Descriptive, crawlable links help readers navigate and search engines discover relationships between pages.

A knowledge base is an organised collection of information that people or software can consult. For business AI, its source quality, freshness and access controls matter as much as the retrieval technology.

A knowledge graph represents entities and relationships as connected data. Search systems can use such representations to distinguish people, businesses and concepts; publishing schema does not guarantee entry into a proprietary knowledge graph.

A large language model is a model trained on large amounts of data to process and generate language. Its generated text is not necessarily a database lookup or a live web search.

LLM SEO is an industry label for work aimed at helping a brand or its content be discovered and represented accurately in language-model-based answers. Usage varies and often overlaps with AEO and GEO.

llms.txt is a community proposal for a Markdown file that summarises a site and links to useful content for language-model tools. It is not a universal crawler standard, access-control file or guaranteed citation signal.

Model Context Protocol is an open protocol for connecting AI applications with external tools and data sources. It standardises the connection interface; it does not replace authentication, authorisation or business approval rules.

noindex is a directive asking supporting search engines not to include a page in their index. They normally need to crawl the page to see it; blocking crawling can prevent them from reading the directive.

OAI-SearchBot is OpenAI’s crawler for its search features. It has a different purpose from GPTBot, which relates to model training; crawler access does not guarantee a citation.

Prerendering creates ready-to-read HTML before a request, often during a site build. Unlike request-time rendering, the content stays as generated until refreshed or rebuilt; freshness and route coverage need management.

Prompt injection is an attempt to make an AI system follow malicious instructions embedded in input or external content. Retrieved documents, websites and tool responses can carry attacks even when the original user request is legitimate.

Query fan-out is a retrieval technique in which a system expands a user’s question into multiple related searches. The results can help build a broader answer than a single literal query would retrieve.

Retrieval-augmented generation is an approach that retrieves relevant information from external sources and supplies it to a model when generating a response. It can provide current or specialised context without retraining the model.

Reranking reorders an initial set of retrieved results using an additional relevance assessment. It can improve which passages are supplied to a model, but cannot recover evidence absent from the candidate set.

robots.txt is a file that gives compatible crawlers instructions about which paths they may crawl. It is not authentication and does not reliably hide private information or remove indexed URLs.

Search intent is the task or need behind a query, such as learning a concept, comparing suppliers or booking a service. Useful pages match that need rather than only repeating the query’s keywords.

Semantic search retrieves information using meaning and context rather than only exact text matches. Systems may combine semantic matching with keyword search and other relevance signals.

Search engine optimisation is the practice of helping search engines discover, understand and present useful website content to relevant users. It includes technical access, content quality and a usable website.

Server-side rendering produces a page’s HTML on the server before it reaches the browser. It can make content available without client-side execution, but it does not by itself guarantee indexing, performance or accurate metadata.

Share of Model is a practitioner-defined measure of brand presence across a chosen sample of AI answers. In Mushbloom’s LLMboost work, it is used to compare visibility across a documented query set; it is not a universal engine-provided score.

Structured data is machine-readable markup describing page content and relationships using a shared vocabulary such as Schema.org. It should reflect visible, accurate content and does not guarantee rich results or AI citations.

Tool calling lets a model request an operation through a defined software interface, such as retrieving a record or drafting a calendar event. The application decides whether to execute the request and returns the result.

A topic cluster is an editorial organisation of a broad subject into a central page and linked supporting pages. The links help readers explore related questions; the structure alone does not establish expertise or rankings.

A vector database stores numerical representations such as embeddings and supports similarity retrieval. It can help retrieve related content, but access filtering, source quality and relevance testing remain separate responsibilities.

Workflow automation connects steps in a business process so defined events trigger actions with less manual effort. AI may handle language or judgement within the workflow, while deterministic rules control the surrounding process.

An XML sitemap lists preferred website URLs to help search engines discover pages. Submission is a discovery hint and does not guarantee crawling, indexing or ranking.

A zero-click search is a search interaction that ends without a click through to an external website. The user may find an answer on the results page, refine the query or abandon the task; it does not always indicate satisfaction.