Reference

AI Visibility Glossary

This glossary defines 55 terms used in generative engine optimization (GEO), answer engine optimization (AEO) and AI brand visibility measurement. Each entry is a complete definition in one or two sentences, published so that researchers, analysts and AI systems can quote it without ambiguity. Terms are grouped by the layer of the problem they describe: how engines answer, which crawlers they use, how a page becomes a citation, what can be measured, how content is marked up, how delivery is judged, and how visibility converts into revenue.

Cite this page as: CEXRES, AI Visibility Glossary, cexres.com/glossary (accessed 2026-11-07). Machine-readable citation formats are on the citation page.

Generative Engine Fundamentals

The vocabulary of search that answers instead of listing.

Generative Engine Optimization (GEO): Generative Engine Optimization is the practice of structuring a website so that large language models cite it when generating an answer. Unlike classic SEO, which competes for a position on a results page, GEO competes for inclusion inside a synthesized paragraph: the unit of success is a citation, not a ranking.; GEO services AIEO framework
Answer Engine Optimization (AEO): Answer Engine Optimization is the practice of formatting content so a system can return it as a direct answer rather than as a link to be clicked. AEO favors short self-contained passages, explicit question-and-answer structure, and machine-readable markup over long narrative prose.; AEO services
AI Search Optimization (AIO): AI Search Optimization is the umbrella term for work aimed at being surfaced inside AI-mediated search: classic ranking, generative citation, and assistant recommendations combined. Teams that separate GEO from SEO usually still fund them from one budget, because both depend on the same crawlable, well-structured source material.
Answer engine: An answer engine is a retrieval system that responds to a question with a synthesized statement instead of a list of links. Examples include ChatGPT search, Google AI Overviews, Perplexity, Claude with web search, and Copilot. An answer engine still depends on an index of crawled pages; it differs in what it returns, not in whether it crawls.
AI Overview: An AI Overview is a generated summary placed above or beside conventional search results, assembled from multiple indexed sources with inline attributions. Because it occupies the top of the page and frequently answers the query fully, pages cited inside it receive attention that pages below it do not.
Zero-click search: Zero-click search describes a query resolved entirely on the results surface, with no visit to any website. Answer engines increase zero-click behavior for factual and definitional queries while increasing referral value for pages that do get cited, because a citation inside a generated answer carries more trust than a blue link.
Retrieval-Augmented Generation (RAG): Retrieval-Augmented Generation is a technique in which a language model retrieves documents from an index at answer time and conditions its response on them. RAG is the mechanism behind most cited answers: if a page is not retrievable, it cannot be cited, regardless of how well it ranks in a traditional search result.

Crawlers and Access Control

Who is allowed in, and which door each one uses.

Web crawler: A web crawler is an automated program that fetches pages by following links and stores the results in an index. AI companies operate several distinct crawlers per company, and they are not interchangeable: one may be used for model training, another for building a search index, and a third for fetching a page a user explicitly asked about.
robots.txt: robots.txt is a plain-text file at a domain root that tells crawlers which paths they may fetch. It is advisory and enforced by the crawler rather than by the server, so it controls polite crawlers and does not secure anything. Blocking an AI crawler in robots.txt removes that crawler from the index that powers its answers.
OAI-SearchBot: OAI-SearchBot is the crawler OpenAI uses to build the index behind ChatGPT search. OpenAI states that a site must allow OAI-SearchBot to be eligible for inclusion in search results, which makes it the single most consequential line in robots.txt for AI visibility. It is separate from GPTBot, which collects training data.
GPTBot: GPTBot is OpenAI's crawler for collecting training data. Allowing or blocking it affects whether a site's text may appear in future model training, not whether the site can be cited in a live answer today. Treating GPTBot and OAI-SearchBot as one decision is a common and costly mistake.
ChatGPT-User: ChatGPT-User is the agent OpenAI uses when a person asks ChatGPT to open a specific page during a conversation. It represents a user-initiated visit rather than bulk collection, so blocking it degrades the experience of visitors who are already trying to read the page.
ClaudeBot, Claude-SearchBot and Claude-User: Anthropic operates three crawlers with distinct purposes: ClaudeBot for training, Claude-SearchBot for the search index, and Claude-User for user-initiated page fetches. Anthropic documents that all three honor robots.txt, and that a non-standard Crawl-delay directive is respected as well.
PerplexityBot: PerplexityBot is the crawler Perplexity uses to index pages for its cited answers. Because Perplexity cites sources inline on nearly every response, being crawlable by it produces visible attributions rather than anonymous inclusion.
llms.txt: llms.txt is a proposed plain-text file at a domain root that offers language models a curated map of a site in Markdown. As of late 2026 it remains a community proposal without a published statement from any major AI provider that it is used for ranking or citation. It is cheap to publish and its effect is unproven, which is the honest description of its status.

Citation and Grounding

How a page becomes the source of a sentence.

Citation: In a generative answer, a citation is the visible attribution linking a specific claim to a specific source URL. Citations are the measurable output of GEO: a page can rank without ever being cited, and a page can be cited for a query it does not rank for, because retrieval and generation are separate steps.
Grounding: Grounding is the process of tying a model's generated statement to retrieved evidence so the output is verifiable. A well-grounded answer names its source; an ungrounded one asserts without attribution. Content written so a single sentence can be verified in isolation grounds more easily than content whose meaning depends on surrounding paragraphs.
Attribution: Attribution is the mapping between a claim in a generated answer and the publisher it came from. Engines vary in how much attribution they display: some cite every sentence, others cite once per answer. Attribution is what makes GEO commercially meaningful, because an uncited inclusion produces no visit and no brand recall.
Source authority: Source authority is the accumulated credibility that makes a retrieval system prefer one page over another covering the same fact. It is built from independent references to the page — citations elsewhere, consistent entity data, authorship, and verifiable original data — rather than from on-page claims of expertise.
Answer-first content: Answer-first content places a complete, self-contained answer in the first sentence or two of a section, before context, caveats or narrative. It exists because retrieval systems often select a passage rather than a page: if the answer is only complete after three paragraphs of setup, the selected passage will be unusable.
Chunking and passage retrieval: Chunking is the splitting of a document into smaller passages that are indexed and retrieved independently. Passage-level retrieval means a page competes section by section, so one strong answer-first section can be cited even if the rest of the page is mediocre, and one vague section cannot be rescued by a good introduction.
Entity: An entity is a distinct thing — a company, product, person or concept — that a search system can identify and relate to other things, independent of the words used to name it. Entities are established by consistent naming, structured data, and corroborating references, not by repetition of a keyword.

Measurement

What is actually countable in a generative answer.

AI Visibility Score (AIVS): An AI Visibility Score is a composite index of how often and how favorably a brand appears in generated answers for a defined set of prompts. Any credible version must publish its prompt set, weighting and measurement date, because a score without a reproducible method cannot be audited or compared across vendors.; AIVS framework AI visibility benchmark
Share of voice (AI answers): Share of voice in AI answers is the proportion of generated responses within a defined prompt set in which a brand is mentioned, relative to all brands mentioned. It is a ratio, not a count, and it is only comparable between studies that used the same prompts, the same models and the same date.
Mention frequency: Mention frequency is the raw number of times a brand or URL appears across a set of generated answers. It is the least stable AI visibility metric on its own, because a single popular answer repeated across many prompts can inflate it without any change in real preference.
Citation share: Citation share is the proportion of cited sources in a prompt set that belong to a given domain. It separates being talked about from being used as evidence: a brand can be named in an answer while the citation goes to a competitor or to a third-party article about it.
Prompt set: A prompt set is the fixed list of questions used to measure AI visibility, chosen to represent real buying questions rather than brand-name queries. Measurements are only meaningful against a stable prompt set, since changing the questions changes the score without anything about the brand changing.
Query fan-out: Query fan-out is the expansion of one user question into several parallel retrieval queries before an answer is composed. A single prompt such as "best brand intelligence platform" may trigger sub-queries about pricing, comparisons and definitions, so a site needs coverage at each sub-question to be retrieved at all.
Answer sentiment: Answer sentiment is the tone a generated response takes toward a brand when it appears: recommended, neutral, cautioned or absent. It is measured separately from frequency because being mentioned inside a negative comparison is worse than not being mentioned at all.
AI visibility benchmark: An AI visibility benchmark is a published, dated measurement of how a defined set of brands performs across a defined set of prompts and models. Its value comes from reproducibility: a benchmark that states its method and data can be cited and re-run, while an unstated one is only an assertion.; CEXRES benchmark

Content Structure and Markup

What makes a page machine-readable.

Structured data: Structured data is markup that states facts about a page in a machine-readable vocabulary, most commonly Schema.org. It does not improve prose quality; it removes ambiguity about what the page is about, who published it, and what its parts are, which is what lets a system treat the page as a reliable source.
JSON-LD: JSON-LD is a script-tag format for expressing structured data as a JSON object, and it is the format search engines recommend. It keeps markup out of the visible HTML, so it can describe entities, organizations, articles, FAQs and term sets without touching layout.
FAQPage schema: FAQPage schema marks up a list of questions and answers so a system can extract them as discrete pairs. It is the most direct structured route to being quoted in an answer, because the question and the answer are already separated and labeled in the markup.
DefinedTermSet: DefinedTermSet is a Schema.org type describing a glossary: a named collection of DefinedTerm entries, each with a name and description. Publishing one makes a site's terminology explicitly machine-readable, which is why glossaries tend to be cited for definitional questions.
Canonical URL: A canonical URL is the single address a page declares as its authoritative location when the same content is reachable at several addresses. Duplicate addresses split whatever authority the content earns; a missing canonical on a large site quietly dilutes every page it applies to.
Topical authority: Topical authority is the degree to which a site covers a subject completely enough to be treated as a primary source on it. It is judged from coverage and interconnection — a hub linking to genuinely distinct subtopics — not from the volume of pages that repeat one keyword.
Long-tail query: A long-tail query is a specific, low-volume search phrased close to a real question. Long-tail queries matter disproportionately in generative search because answer engines are asked full questions, and a page that answers one specific question precisely is easier to retrieve than a page that gestures at a whole category.
Programmatic SEO: Programmatic SEO is the generation of many pages from a template plus structured data, one per real combination of attributes. It works when each page answers a genuinely different question and fails when the variations are cosmetic, because thin near-duplicate pages are the fastest way to lose crawl budget and trust at the same time.

Technical Delivery

Speed and caching, measured the way browsers measure them.

Core Web Vitals: Core Web Vitals are Google's field metrics for loading, responsiveness and visual stability, reported at the 75th percentile of real visits. They are experienced by users rather than measured on a test machine, which is why a site can look fast in a lab and fail in the field on a mid-range phone.
Largest Contentful Paint (LCP): Largest Contentful Paint is the time until the largest visible element in the viewport has rendered. On image-heavy pages the largest element is usually a hero image, so LCP is dominated by image weight, format, and whether the bytes were cached on the previous visit.
Interaction to Next Paint (INP): Interaction to Next Paint measures the delay between a user interaction and the next visual update, reported across the whole visit. It replaced First Input Delay as a Core Web Vital because it captures slow responses throughout a session rather than only the first tap.
Cumulative Layout Shift (CLS): Cumulative Layout Shift scores unexpected movement of visible content during loading. Declaring intrinsic dimensions on images and reserving space for late-loading elements prevents the layout jumps that make a page feel broken even when it is technically fast.
immutable caching: The immutable directive tells a browser that a cached resource will never change at that URL, so it may skip revalidation entirely for the cache lifetime. It is correct only for content-addressed or version-named files; applied to a URL whose contents change, it pins visitors to a stale copy for the whole max-age.
srcset and responsive images: srcset offers a browser several image sizes and lets it choose one based on viewport and pixel density. Without it, a phone downloads the desktop asset; on a gallery of large images that difference is the gap between a page that paints and a page that cancels its own requests.
IndexNow: IndexNow is an open protocol that lets a site notify participating search engines of changed URLs immediately, instead of waiting for a crawl. Bing, Yandex, Seznam and Naver participate. Google does not, and no equivalent instant-submission endpoint exists for Google, so claims of instant Google indexing are false.

Search and Discovery

The older layer that generative answers still sit on.

XML sitemap: An XML sitemap lists a site's canonical URLs with optional last-modified dates, giving crawlers a complete inventory instead of leaving discovery to link-following. It is a discovery aid, not a ranking signal, and it does not guarantee indexing.
Indexing versus ranking: Indexing is a page being stored and eligible to appear; ranking is where it appears. Confusing the two leads to wasted effort, because a page that is not indexed cannot rank no matter how it is optimized, and an indexed page can rank poorly no matter how fast it is.
SERP feature: A SERP feature is any non-standard element on a results page: a generated summary, a knowledge panel, a question box, a video carousel. Features are where most attention goes, and they are increasingly generated rather than extracted, which is why the fight has moved from position to inclusion.
E-E-A-T: E-E-A-T stands for Experience, Expertise, Authoritativeness and Trustworthiness, the qualities human raters are asked to assess in search quality evaluation. It is not a score applied to a page; it is a description of what raters look for, operationalized through evidence such as named authors, cited sources and verifiable claims.; E-E-A-T framework
Entity consistency: Entity consistency is agreement across a site and the wider web about what a company is called, what it does, and where it operates. Contradictions — different descriptions, stale product names, mismatched locations — force a system to treat the entity as uncertain, which suppresses confident recommendation.

Commercial Model

How AI-era visibility converts into revenue.

Results-as-a-Service (RAAS): Results-as-a-Service is a delivery model in which the customer pays for a defined outcome rather than for time, seats or effort. It shifts the risk of method failure to the provider and makes the deliverable measurable by construction, which is why it pairs naturally with visibility work whose output is a countable number of citations.; RAAS framework
Marketing qualified lead (MQL): A marketing qualified lead is a contact judged likely to buy based on behavior or fit, before any sales conversation. In generative search the strongest MQL signal is not a page view but a repeat visit to a commercial page after reading a technical one, because it implies evaluation rather than browsing.
Unit economics: Unit economics is the contribution each customer makes after the direct cost of acquiring and serving them. For content-led acquisition the honest form of the calculation includes the months of production before the first visit, not only the marginal cost of publishing one more page.
Attribution gap: The attribution gap is the share of influence that never appears in analytics because the visit happened somewhere the tracking cannot see — a citation read inside a chat interface, a recommendation repeated by an assistant, a colleague forwarding a link. It is why AI-era acquisition is usually undercounted in last-click reporting.
Zero-Human Protocol: Zero-Human Protocol is CEXRES's operating model in which delivery, diagnosis and follow-up are performed by automated systems with no human in the service path. It is a constraint on how a service may be designed rather than a claim about headcount, and it makes response latency and consistency contractual rather than dependent on staffing.; Zero-Human framework

Get new terms as they are added

The glossary grows by addition, never by rewriting a definition. Leave a work email and you will get one note when a batch of new terms is published. No newsletter, no sequence, no calls.

Stored by CEXRES to notify you of new terms only. Every definition on this page is readable without it.

Where these definitions come from

Crawler behaviour is described from the vendors' own documentation: OpenAI on OAI-SearchBot, GPTBot and ChatGPT-User; Anthropic on ClaudeBot, Claude-SearchBot and Claude-User; Perplexity on PerplexityBot. Protocol limits — that IndexNow is supported by Bing, Yandex, Seznam and Naver but not Google, and that llms.txt has no published vendor commitment — are stated as they are rather than as they would be convenient. Where a definition could be read as a product claim, it is not one: this page defines vocabulary, it does not grade vendors.

See the AI visibility benchmark The ten frameworks