Short answer
A mention is not a citation, a citation is not a visit and a visit is not a lead. Combining them into one “AI visibility” number makes the report difficult to audit and easy to misread.
The cleanest model keeps the answer layer, source layer, website layer and business layer separate. Teams can then connect records when evidence exists without pretending every step is observable.
Four signals that need different records
| Signal | Definition | Minimum evidence |
|---|---|---|
| Brand mention | The captured answer names the brand. | Full answer capture, exact query, platform and date |
| Cited mention | The answer names the brand and displays a source link or citation associated with the response. | Answer capture plus visible source destination |
| AI referral | A human website session arrives with an observable AI-assistant referrer or campaign marker. | Session timestamp, referrer or UTM, landing page and bot exclusion |
| Attributed lead | An enquiry record is connected to a known session or declared source under a stated attribution rule. | Lead timestamp, session or campaign evidence, qualification status and rule used |
A crawler request is an operational access event. It is not a person, session, referral or conversion.
One journey, four possible records
A buyer researching a local provider
An AI assistant names three companies and displays two source links. That produces one answer capture with three mentions and two visible citations. The buyer clicks one source, creating a referral session. After returning directly the next day, the buyer submits a form.
The report should not collapse this into “one AI conversion” without explanation. It can state that the first known session came from an AI referrer and the enquiry occurred on a later direct session, if the consented first-party data supports that sequence.
Recommended event fields
| Layer | Fields |
|---|---|
| Answer observation | query_id, platform, market, language, captured_at, brand_mentioned, cited, source_urls |
| Website session | session_id, started_at, landing_page, document_referrer, source, medium, bot_class |
| Lead | lead_id, submitted_at, session_id when available, declared_source, qualification_status |
| Attribution | model, first_known_source, last_known_source, confidence, notes |
A reporting template that stays readable
- Scope: platforms, markets, languages and dates.
- Query set: total fixed questions and any approved changes.
- Answer results: mentions, cited mentions and cited domains.
- Human activity: observable AI-referral sessions and engaged sessions.
- Business activity: enquiries, qualified leads and attribution rule.
- Limitations: missing referrers, cross-device journeys, answer variability and sample size.
How to calculate rates without changing the denominator
Mention rate is the number of captured answers containing the brand divided by all eligible captured answers in the fixed set. Cited-mention rate uses answers where the brand is both named and visibly sourced. Referral conversion rate uses human referral sessions as its denominator, never query captures or crawler hits.
Publish the count beside every percentage. “4 of 20 captured answers” is more useful than “20% visibility” because the reader can see the sample size.
QA checklist
- Every answer capture has an exact question, date, market and platform.
- Mentions and citations are separate booleans.
- Source URLs are opened and checked rather than inferred from the brand name.
- Known bots are classified before session reporting.
- Direct traffic remains direct when the source is unavailable.
- Lead attribution states first-touch, last-touch or another rule explicitly.
- Counts and denominators appear beside rates.
Primary analytics references
Google Analytics documentation distinguishes source and medium fields and explains that direct traffic is used when clear referral information is unavailable. The same caution applies to AI referrals: do not manufacture a source when the browser or platform did not provide one.
Where to go next
Use the Nexus AI Visibility Framework to place this topic inside a complete measurement system, or review the AI Visibility Audit scope.
13 Jul 2026 — Expanded with an independent structure, examples, implementation guidance and primary references.