AI Visibility Reports: Connect Mentions, Citations, Referrals and Revenue

An AI visibility report should keep mentions, citations, referrals and revenue in separate ledgers, then connect them with transparent identifiers. Combining them into one score hides which stage changed and whether the movement helped the business.

Define the four layers

Layer	Unit	Primary evidence
Mention	Answer names entity	Saved answer and prompt
Citation	Answer links to source	Source URL plus supported claim
Referral	Visit with detectable origin	Analytics session and landing page
Revenue	Qualified business outcome	Conversion system with value

A mention is not automatically a citation; a citation is not a visit; a visit is not revenue. Preserve the funnel rather than implying a direct conversion path you did not measure.

Build a frozen observation design

Store exact prompt, platform, model or mode when available, date, locale, device/account state, answer and source list. Use a stable prompt set for trend comparisons. Report completed, failed and unavailable rows. Randomly spot-check or fully validate whether cited sources support nearby claims.

Our measurement crosswalk maps differing vendor terms.

Use platform-native data where available

Bing introduced AI Performance reporting with citations, cited pages and sampled grounding queries in Webmaster Tools; read the official announcement for its scope. Google’s generative AI performance reports expose impressions, pages, countries, devices and dates during the rollout, but not grounding queries. OpenAI documents a referral campaign parameter for ChatGPT.

Do not force these units into direct cross-platform rankings. Preserve platform, report name, definition and export date.

Reconcile analytics without claiming completeness

Create channel rules for known AI referrers and campaign parameters, then inspect landing pages and qualified actions. Referrer loss, apps, privacy settings and copy-paste behavior create dark traffic. Report detected referrals as observed, and keep self-reported or modeled influence separate.

Use our GA4 AI referral guide for the implementation.

Connect ledgers with stable keys

Use page canonical, prompt ID, observation ID, session or campaign key where available, conversion ID and date window. Avoid joining an aggregate visibility score to total revenue and implying causation. If attribution is modeled, show assumptions and sensitivity ranges.

A useful report can say: “This page was cited in 6 of 40 completed tracked prompts, received 12 detected referrals and produced two qualified trials.” It should also say what it cannot connect at user level.

Show decisions and uncertainty

• Which high-value pages are eligible but absent from observed citations?
• Which cited pages receive no detected referrals?
• Which referral pages fail to convert?
• Which prompts changed because the platform or sample changed?
• Which rows are unknown rather than zero?

Pair every recommendation with evidence, an owner, a validation method and a review date.

Avoid dashboard theater

Do not lead with a composite score whose weights can change. Do not hide failed prompts. Do not count source-list appearances without claim validation. Do not show percentages without denominators. Do not claim revenue attribution from temporal coincidence.

Download the AI visibility four-ledger report. Replace the example rows and preserve the layer, unit, denominator, confidence and decision fields. Those fields make the report useful when products and vendor metrics change.

Work one row from observation to decision

Suppose a fixed sample contains 40 prompts. Thirty-two complete successfully, eight fail or are unavailable, and six of the completed answers cite one canonical page. Report 6 of 32 completed observations (18.75%), not 6 of 40 without explanation and not “19% visibility” as though the sample represented the whole platform. If analytics records 12 detectable visits to that page and two qualified trials, keep those as separate referral and outcome counts.

The next action depends on the missing stage. If the page is cited but receives no visits, inspect the answer context and the promise around the source link. If visits arrive but qualified actions do not, inspect page-task fit and conversion friction. If the page is eligible but absent from all valid observations, compare its evidence and intent coverage with cited sources before rewriting it. Record prompt-set changes, product updates, seasonality, and measurement gaps beside the row so the next period is comparable.

For the broader operating model behind these ledgers, return to the AI SEO guide and task hub.