AI Visibility Reporting Glossary
A practical reference for teams measuring how brands appear in AI-generated answers.
AI-search reports can look precise while measuring different things. One dashboard may count a brand mention, another a linked citation, and another a recommendation. This glossary gives content, SEO, analytics, and communications teams a shared vocabulary for reviewing results and making changes.
The definitions are intentionally operational: each term includes what to record or how to use it. For implementation support and deeper AI-search audits, visit Corank.
Answer coverage
Definition: The share of sampled prompts for which an AI system returned a usable answer rather than an error, refusal, empty result, or irrelevant response.
Record: Covered prompts divided by eligible prompts, with the engine, interface, market, and date. Do not treat missing answers as proof that a brand is invisible.
Brand mention
Definition: Any explicit appearance of the tracked brand name or a clearly identified product in the answer text.
Record: The exact text span and the normalized entity it refers to. A mention is not automatically a recommendation or a citation.
Linked citation
Definition: A clickable source reference attached to, or presented alongside, an AI answer.
Record: The raw destination URL exactly as shown, its position, the associated claim if visible, and the resolved canonical URL. Keep linked citations separate from unlinked mentions.
Unlinked source reference
Definition: A source, publication, company, or study named in the answer without a clickable destination.
Record: The quoted reference and the entity it appears to identify. Do not silently convert it into a URL during analysis.
Citation rate
Definition: The percentage of eligible answers that contain at least one citation to the tracked domain or page set.
Formula: Answers with a qualifying citation divided by eligible covered answers.
Caution: State whether multiple citations in one answer count once or many times.
Citation share
Definition: A brand or domain's portion of all qualifying citations within a defined prompt set and competitor cohort.
Record: The numerator, denominator, cohort, engines, and deduplication rule. A domain with many URLs can otherwise appear stronger simply because the report counts duplicates.
Recommendation strength
Definition: The degree to which an answer endorses, ranks, shortlists, or directs a user toward a brand.
Suggested labels: Not mentioned, contextual mention, considered option, recommended, strongest recommendation.
Caution: Keep this separate from sentiment. A positive description is not always a buying recommendation.
Answer position
Definition: Where a brand first appears in the visible answer.
Record: Use a consistent scheme such as first named option, ordinal list position, or paragraph number. Do not mix those schemes in one trend line.
Source position
Definition: Where a cited source appears in the citation list or answer interface.
Caution: Citation interfaces differ by engine and can change. Store a screenshot or raw response with the observation.
Raw cited URL
Definition: The exact URL returned by the AI interface before redirects, tracking removal, or canonicalization.
Why it matters: It preserves evidence and exposes redirect or parameter patterns that normalization may hide.
Resolved URL
Definition: The final destination after following permitted redirects.
Record: Redirect chain, HTTP status, retrieval timestamp, and any login or geographic restriction.
Canonical URL
Definition: The preferred normalized URL for grouping equivalent pages, based on declared canonicals and a documented fallback policy.
Caution: Never discard the raw cited URL. Reports should be able to show both raw evidence and the canonical group.
Citation destination quality
Definition: A practical assessment of whether the cited page is accessible, relevant, current, and useful for the query.
Suggested checks: Public access, topical match, clear authorship, current evidence, readable text, and a logical next action.
Source role
Definition: The function a source appears to serve in an answer.
Suggested labels: Definition, evidence, comparison, product documentation, news, directory, review, community discussion, or primary research.
Caution: Role is an analyst classification unless the AI interface states it directly.
Claim-source alignment
Definition: The degree to which a cited page actually supports the nearby claim.
Suggested labels: Direct support, partial support, contextual only, contradicted, or unclear.
Record: The answer claim and the supporting page passage. A citation next to a claim does not prove alignment.
Claim-source ledger
Definition: A table connecting each material brand claim to its owner, public source, supporting passage, last review date, and status.
Use: It helps teams repair unsupported marketing language before measuring whether AI systems repeat it.
Entity consistency
Definition: Agreement in a brand's public identity across first-party pages and important profiles.
Check: Official name, domain, logo, description, product naming, leadership, contact details, and legacy names.
Prompt inventory
Definition: The approved set of questions used for repeatable monitoring.
Record: Prompt ID, exact wording, intent, audience, market, language, topic, funnel stage, and owner.
Prompt version
Definition: A dated snapshot of the prompt inventory and its eligibility rules.
Why it matters: Changing prompt wording or mix can change visibility results even when the public web has not changed.
Prompt drift
Definition: A material change in the prompt set, user intent, market mix, or eligibility policy that makes two reporting periods less comparable.
Action: Run an overlap or bridge sample before replacing the old benchmark.
Sampling budget
Definition: The number of prompt-engine observations a team can reliably collect and review within a reporting period.
Use: Allocate more samples to high-value topics and uncertain findings, while preserving a stable core benchmark.
Engine surface
Definition: The specific product and interface used to collect an answer, such as a chat interface, search mode, or API.
Caution: Do not merge observations from different surfaces without labeling them.
Retrieval timestamp
Definition: The date and time an answer and its sources were collected, including time zone.
Why it matters: AI answers, indexes, and cited pages change. A URL checked later may not match the original evidence.
Evidence bundle
Definition: The material needed to reproduce or audit an observation.
Minimum contents: Prompt, engine surface, timestamp, raw answer, cited URLs, screenshot or export, normalization output, and analyst notes.
Confidence label
Definition: A transparent statement about how strongly the evidence supports a conclusion.
Suggested labels: Exploratory, directional, supported, or high confidence.
Caution: Define labels in advance. Do not convert small samples into fake numerical precision.
Eligibility policy
Definition: The written rules that determine which prompts, answers, brands, sources, and citations are included in a metric.
Record: Errors, refusals, duplicates, self-citations, redirects, subdomains, locale differences, and competitor inclusion.
Incident threshold
Definition: A documented rule for when a visibility change requires investigation.
Example structure: Trigger only when the change exceeds a chosen magnitude, persists across repeated samples, and affects a priority topic. Avoid alerts based on one volatile answer.
Repair queue
Definition: A prioritized list of actions derived from observed evidence gaps.
Each item should include: Owner, target URL, issue, supporting evidence, planned change, expected outcome, review date, and status.
Minimum reporting contract
A trustworthy AI-visibility export should preserve:
- Prompt ID and exact prompt text
- Prompt-set version and eligibility policy
- Engine surface, market, language, and retrieval timestamp
- Raw answer and raw cited URLs
- Resolved and canonical URLs
- Brand mention and recommendation-strength labels
- Citation role and claim-source alignment
- Confidence label and analyst notes
- Links to the supporting evidence bundle
A report that keeps only the final score cannot explain why the number changed. Preserve enough evidence to distinguish a real visibility shift from prompt drift, interface changes, URL normalization, or sampling noise.
How to use this glossary
Start with five terms that affect your current dashboard, write their exact rules, and test two analysts on the same ten answers. Resolve disagreements before expanding the sample. The aim is not perfect terminology; it is a measurement process that another person can understand and audit.
This glossary is educational and does not guarantee inclusion in any AI-generated answer. Learn more about AI search optimization and brand visibility at https://corank.ai.