The practical AI analytics glossary.
Short definitions for a measurement category that is still learning how to name what it sees.
A–C
- AI agent analytics
- Measurement of automated systems that inspect or act on a website, organized by actor, authorization, attempted task, completion, and failure.
- AI citation
- A link or explicit source reference included in an AI-generated answer. A citation is an appearance in an answer, not evidence that a person visited the cited site.
- AI crawler
- An automated system that discovers or retrieves web content for indexing, training, answer retrieval, monitoring, or another declared purpose.
- AI referral
- A human website visit with observable referral evidence from an AI assistant or answer engine.
- AI referral analytics
- Measurement of observed human visits from AI providers, including landing pages, behavior, conversions, and value.
- AI search visibility
- Observed presence and representation in a defined sample of AI answers, including mentions, citations, linked pages, and factual claims.
- Answer engine
- A search or assistant product that synthesizes a response to a question, often with links or citations, rather than returning only a list of pages.
- Attribution
- A rule or model that connects an observed outcome to one or more preceding sources or interactions. Attribution describes evidence under a method; it does not prove sole causation.
- Canonical fact
- A maintained, authoritative statement about a product or organization, such as pricing, availability, a policy, or specification.
- Citation rate
- The share of monitored answers that explicitly cite or link to the site being measured.
D–M
- Direct traffic
- Visits for which the analytics system has no usable referring source. Direct does not necessarily mean a person typed the address; it can also include lost referral information.
- Evidence chain
- The reviewable sequence connecting a conclusion to observations, such as referral host, first pageview, session actions, and outcome.
- First-party analytics
- Measurement collected within the relationship between a site and its visitor, under the site operator’s control, rather than through a cross-site data relationship.
- Human referral
- A visit made by a person following a link from another source. It should be reported separately from automated requests by crawlers or agents.
- Landing page
- The first measured page of a visit. For AI referrals, it often reveals the question, comparison, or task that preceded the click.
- Meaningful action
- An on-site behavior tied to genuine intent, such as viewing pricing, reading key documentation, starting a trial, submitting a lead, or completing a purchase.
- Mention rate
- The share of monitored AI answers in which the measured brand, product, or entity appears under a defined matching rule.
- Modeled attribution
- An estimated connection between interactions and outcomes based on a documented model. It should remain distinguishable from directly observed referral evidence.
- Monitored answer
- An AI-generated response captured for a defined question, provider, locale, and time as part of a repeatable visibility sample.
N–Z
- Observed
- Supported directly by the available measurement evidence, within its stated coverage. Observed does not mean complete or causal.
- Outcome
- A defined result that matters to the organization, such as a qualified lead, trial, purchase, renewal, or another verified goal.
- Presence monitoring
- Repeated observation of brand mentions, citations, representation, and factual accuracy across a defined AI answer set.
- Referrer
- Source information a browser may send when a person follows a link. Referrer availability depends on applications, redirects, browser policy, and privacy controls.
- Retrieval bot
- An automated system that fetches current web content so an assistant or answer engine can use it when responding.
- Session
- A bounded group of related interactions under a stated timeout and continuity rule. Sessions are an analytics convention, not a natural fact.
- Source taxonomy
- A maintained set of rules that maps observed hosts, campaign parameters, and other evidence into consistent acquisition categories and providers.
- Task completion rate
- The share of attempted agent tasks that reach the explicitly defined successful result.
- Unknown
- A deliberate category used when the evidence cannot support a more specific classification. Unknown should not be silently converted to zero or assigned to a convenient source.
- User-delegated agent
- An automated system acting on behalf of a person to complete a permitted task, ideally with evidence of identity, delegation, and authorization.
Read the field notes.
Wandered is in private development