Learn
August 12, 2026 · 10 min read
By Marcus Bransbury · Founder, Robot Visible
AEO glossary: 45 AI visibility terms
Clear definitions of 45 AEO and AI visibility terms, with stable anchors and links to the canonical guide for every technical or measurement concept.
How to use this glossary
This is the stable vocabulary used throughout Robot Visible's AEO resource hub. Each definition has a permanent fragment link and one contextual route into the guide that owns the deeper explanation. Link to the individual term when a report, brief or implementation note needs an unambiguous definition; follow its guide link when the reader needs method, evidence or controls.
The list is intentionally narrower than a general search or machine-learning dictionary. It covers the terms a publisher needs to decide whether a page is accessible, compare what competing sources provide, record what an answer actually did, and measure a change without turning inference into fact.
Evidence terms stay separate
Eligible, Competitive and Observed are three evidence states, not levels of one score. Eligible evidence concerns whether a page can be reached and interpreted. Competitive evidence concerns the relative source and answer landscape. Observed evidence records a real answer from a named provider and surface. Passing an eligibility check cannot prove a citation, and absence from one answer cannot prove a technical failure.
Mention, recommendation and citation are also distinct answer outcomes. A brand can be named without being put forward, recommended without an owned page being cited, or cited as a factual source without being recommended. Reporting them independently preserves the next action; a blended visibility score can hide it.
Stable anchors and canonical guides
Every entry below has a stable lowercase fragment such as #citation-rate. The page also publishes a Schema.org DefinedTermSet whose DefinedTerm descriptions come from the same records as the visible definitions. The markup describes the glossary; it does not create an AI-search feature, ranking signal or guaranteed citation.1,2
A term's guide link identifies the best current page for its full treatment, not the only page allowed to use the term. The same canonical guide may own several related definitions. That creates useful internal-link destinations without producing thin pages for individual words or pretending that every industry label has settled into one universal meaning.
Maintenance and interpretation
Crawler roles, robots behaviour, search features and reporting surfaces can change, so their linked references carry dates, source notes and limitations. The robots.txt definition follows the Robots Exclusion Protocol's role as a crawler request standard rather than a security control. Provider-specific implementation belongs in the maintained crawler and engine guides, where it can be reviewed without silently changing this shared vocabulary.3,4,5
Definitions describe how Robot Visible uses a term in evidence and product reporting. Where the market uses a word more broadly, the definition states the boundary that matters here. Provider errors remain unknown; an unchecked provider is not a miss; a captured answer is a dated sample; and a change followed by movement is not automatically proof of causation.
Definitions A–Z
Use a letter to jump through the list. Each definition has a permanent link and a route to the guide that explains the concept in context.
- Answer engine optimisation (AEO)
Answer engine optimisation (AEO) is the practice of making a business the answer AI assistants give when a buyer asks for a recommendation. AI visibility is the outcome it is measured by.
Read the canonical AEO guide.
- AI readiness
AI readiness is Robot Visible's 0–100 assessment of whether a page is readable, permitted and answerable from its initial HTML. It is eligibility evidence, not a prediction of ranking, recommendation, citation or traffic.
Read the AI-readiness assessment guide.
- AI search
AI search is the channel in which a search or assistant product retrieves information and generates, summarizes or organizes an answer. The term names a surface category, not a single engine, crawler or measurement method.
Read the measurable AI surfaces guide.
- AI visibility
AI visibility is the measured outcome of how a brand, entity or website appears across defined questions and answer surfaces. It must be reported with explicit metrics such as mentions, recommendations and citations rather than inferred from technical readiness.
Read the AI visibility methodology.
- Answer engine
An answer engine is a product or surface that returns a composed answer to a question, often using search, retrieval, model knowledge or tools. Different answer engines expose different citations, controls, locales and levels of reproducibility.
Read the answer-source selection guide.
- Baseline
A baseline is the dated evidence collected before a defined website change. A useful baseline preserves the page, query, provider, surface, locale, result and error state needed to make a later comparison interpretable.
Read the citation measurement guide.
- Canonical URL
A canonical URL is the preferred public URL for a page or substantially duplicate set, declared through consistent links, redirects, sitemap entries and an appropriate canonical tag. It guides consolidation but does not guarantee indexing or citation.
Read the technical AEO checklist.
- Change cycle
A change cycle is the evidence chain connecting a frozen baseline, diagnosed constraint, approved intervention, publication record, verification and comparable follow-up measurement. It records what happened without treating sequence as proof of causation.
Read the AEO Action Loop playbook.
- Citation
A citation is an answer-engine attribution to a source URL that supports or accompanies a generated response. Robot Visible counts it only when the captured evidence retains the query, provider, surface, source URL and timestamp.
Read the citation evidence methodology.
- Citation rate
Citation rate is the share of valid captured answers that cite at least one URL on the measured domain. Its denominator excludes provider errors and missing evidence, and the result applies only to the defined query panel and surfaces.
Read the citation tracking guide.
- Citation retention
Citation retention is the share of previously cited questions that continue to cite the measured domain in later comparable valid runs. Several observations are usually needed because generated answers and source selection vary normally.
Read the longitudinal measurement guide.
- Competitive evidence
Competitive evidence shows how a page's relevance, factual coverage, proof, freshness, clarity or corroboration compares with sources selected for the same buyer question. It does not establish that the page was observed in an answer.
Read the competitive source guide.
- Corroboration
Corroboration is independent evidence that supports an identity or claim, such as a regulator, partner, customer, marketplace or primary source. Repetition on publisher-controlled pages is consistency, not independent corroboration.
Read the trust and corroboration guide.
- Crawler
A crawler is automated software that requests web resources for a declared or inferred purpose such as search indexing, model training or product validation. A user-agent string alone does not prove a request is genuine or reveal later use.
Read the maintained crawler reference.
- Entity
An entity is a distinguishable real-world thing—such as an organization, product, person, place or service—whose identity can be reconciled across names, URLs, facts and relationships. Clear entity data reduces ambiguity but does not confer authority.
Read the entity trust guide.
- Eligible evidence
Eligible evidence shows that an intended system could access and interpret a resource under the tested conditions. Examples include a successful fetch, index eligibility and readable initial HTML; none proves selection, mention or citation.
Read the eligibility testing guide.
- Generative engine optimisation (GEO)
Generative engine optimisation is a secondary industry term for work intended to improve representation in generated answers. Robot Visible uses AEO as the category and AI visibility as the measured outcome so tactics and evidence do not blur together.
Read the AEO, GEO and AI visibility comparison.
- Grounded answer
A grounded answer is a generated response informed by retrieved or supplied source material rather than model parameters alone. Grounding can improve currency and traceability, but it does not guarantee that every statement has a visible citation.
Read the answer-surface measurement guide.
- Grounding
Grounding is the process of supplying retrieved documents, data or tool results as context for a generated response. Its implementation, source display and publisher controls vary by provider and surface, so the term should not imply one crawler path.
Read the retrieval and source-selection guide.
- Index eligibility
Index eligibility means a search system is permitted and technically able to consider a URL for its index under the observed directives and response. Eligibility is not the same as confirmed indexing, ranking, retrieval or citation.
Read the website discovery diagnosis.
- Initial HTML
Initial HTML is the document body delivered in the first successful page response before client-side JavaScript changes the DOM. Keeping essential meaning there reduces dependence on rendering capabilities and interaction timing.
Read the initial-HTML readiness guide.
- Internal link
An internal link is a crawlable link from one page on a site to another, ideally using descriptive anchor text and a real href. It helps readers and crawlers discover relationships but does not guarantee the destination is indexed or selected.
Read the site discovery checklist.
- llms.txt
llms.txt is an emerging plain-text convention for pointing language-model consumers toward useful site content. It is optional, has no universal provider adoption, and cannot replace crawlable pages, sitemaps, visible evidence or ordinary search eligibility.
Read the structured data and llms.txt guide.
- Mention
A mention occurs when a valid captured answer names the measured brand, organization, product or entity. It can be positive, neutral or adverse and may appear without an owned citation, recommendation or attributable visit.
Read the mention and citation methodology.
- Mention rate
Mention rate is the percentage of valid captured answers in a defined panel that name the measured entity. Provider failures are excluded from the denominator, and the metric says nothing by itself about sentiment, position or source.
Read the AI visibility measurement guide.
- noindex
noindex is a robots directive requesting that compliant search engines exclude a resource from their indexes. The crawler generally must be allowed to fetch the resource to see the directive, and removal may not be immediate.
Read the crawler and indexing controls reference.
- Observed evidence
Observed evidence is a captured provider result showing what happened for a specific query, surface and time, including the answer and cited source URLs where present. It is bounded evidence, not a permanent ranking claim.
Read the observed evidence methodology.
- Provider
A provider is the organization or service responsible for the measured answer capability, such as Google, OpenAI, Perplexity, Anthropic or Microsoft. Provider identity should be recorded separately from model and surface.
Read the provider and surface guide.
- Provider error
A provider error is a failed or invalid measurement caused by timeout, access, quota, authentication, service or parsing conditions. It remains unknown and must not be converted into an absent mention, lost citation or zero score.
Read the evidence-state methodology.
- Query
A query is the exact question or instruction submitted to a search or answer surface for measurement. Store its wording without silent normalization because small changes can alter retrieval, intent and source selection.
Read the query measurement guide.
- Query fan-out
Query fan-out is a retrieval technique in which a system issues multiple related searches or subqueries while building one response. Publishers usually cannot see every generated subquery, so it is not a template for mass-producing pages.
Read the Google AI sourcing guide.
- Query panel
A query panel is a maintained set of real customer questions used for comparable measurement. It normally contains a stable core, explicit intent and importance, controlled settings where possible, and documented additions or retirements.
Read the query-panel design guide.
- Recommendation
A recommendation occurs when an answer presents an entity as a suitable option, not merely when it names or cites it. The classification should preserve context and polarity because warnings and exclusions are not positive recommendations.
Read the recommendation selection guide.
- Referral
A referral is an observable visit arriving through a link or tagged source from another product or website. It measures a traffic event, not all no-click visibility, copied links, cross-device journeys or the answer that influenced the visit.
Read the AI referral measurement guide.
- Rendered HTML
Rendered HTML is the document state after a browser executes allowed scripts and updates the page. Comparing it with initial HTML can reveal missing or contradictory content, but one browser render does not reproduce every crawler.
Read the initial and rendered HTML guide.
- Retrieval
Retrieval is the process of finding and selecting documents or data relevant to a question before or during answer generation. Crawl access is one prerequisite; relevance, indexing, freshness and source competition affect what is returned.
Read the retrieval pipeline guide.
- robots.txt
robots.txt is a host-level file expressing advisory crawl rules for named user agents and paths. It is not authentication, does not remove an already indexed URL by itself, and should be tested for each intended crawler role.
Read the robots.txt crawler reference.
- Sitemap
A sitemap is a machine-readable list of canonical URLs a publisher wants search systems to discover, often with honest modification dates. Submission aids discovery but does not guarantee crawling, indexing, ranking or citation.
Read the sitemap and discovery checklist.
- Source URL
A source URL is the exact web address attributed or linked as evidence in a captured answer. Preserving it is essential because a domain-level count cannot show which page, passage or publisher claim supported the response.
Read the source-level tracking guide.
- Structured data
Structured data is machine-readable markup that labels entities, properties and relationships already supported by visible content. It can reduce ambiguity and enable named features, but it is not special AI markup or proof of a citation.
Read the structured data guide.
- Surface
A surface is the specific product experience where an answer was observed, such as Google AI Mode, ChatGPT search, a provider API or a consumer assistant. Results from one surface must not be relabeled as another.
Read the answer-surface reference.
- Training crawler
A training crawler collects content for declared model-development purposes rather than automatic search discovery. Where providers offer separate controls, declining training need not require blocking their distinct search crawler.
Read the crawler-role reference.
- User-triggered fetch
A user-triggered fetch is a request made in response to a person's action, such as asking an assistant to open or summarize a URL. Providers may document different robots behaviour and IP ranges from automatic search crawling.
Read the user-fetch crawler reference.
- Visibility action
A visibility action is a specific, owned change intended to address a diagnosed eligibility, competitive or measurement constraint. It should record approval, completion evidence, expected outcome, verification and follow-up without promising causation.
Read the visibility action playbook.
Sources and further reading
- DefinedTermSet — Schema.org. Defines a set of words, names, acronyms, phrases or other terms with agreed meanings, and the hasDefinedTerm relationship used on this page.
- DefinedTerm — Schema.org. Defines a term contained in a glossary or classification scheme and the inDefinedTermSet relationship used for each visible entry.
- Robots Exclusion Protocol — Internet Engineering Task Force. Standardizes robots.txt crawler access requests and explicitly states that the protocol is not a substitute for valid content security measures.
- Google's guide to optimizing for generative AI features — Google Search Central. Documents established search fundamentals, crawl access, visible content, structured-data agreement and measurement for Google's AI features without special AI markup.
- Publishers and Developers FAQ — OpenAI. Documents OpenAI's separate search, training and user-triggered agents, publisher controls, public-page eligibility and referral reporting.
Continue learning
See where your website stands
Run a free scan and get your AI readiness score across all six categories, with the gaps to fix first.