TL;DR
A mention is an answer observation. A citation points to a source. A visit records website activity where measurable. A request or appointment needs its own evidence. Define these events before calculating a conversion rate or attributing business results to AI visibility.
Build a measurement dictionary
Ask whoever prepares the report to complete one row for each number used in decision-making:
| Event | What it establishes | What it does not establish |
|---|---|---|
| Practice mention | The practice appeared in an observed answer | A recommendation, visit or booking |
| Website citation | A page was referenced as a source | That the practice was endorsed |
| Measured website session | Activity recorded under the analytics setup | A unique prospective patient in every case |
| Appointment-link click | Someone activated that link | A successful scheduler load or submission |
| Appointment request | A request was received | A confirmed appointment |
| Confirmed appointment | A scheduling outcome was verified | Attendance, treatment or revenue |
Add the data source, reporting period, definition and known limitations beside each row. This shared dictionary prevents a dashboard label from quietly changing meaning between meetings.
Do not build a funnel from unrelated totals
When a report shows totals for AI citations, measured referral sessions and confirmed appointments, check what each total includes. Dividing all appointments by all citations does not, by itself, produce a valid AI booking rate.
That calculation is not justified unless the data establishes a compatible relationship between those events. The citations may cover different surfaces, and the appointments may have come from other sources. In Google Analytics, a session is a period of interaction, and a user can start later sessions; session totals therefore do not count unique prospective patients. Google’s session documentation explains those definitions. A mathematically correct division can still describe the wrong thing.
Report the event totals separately until an appropriate method connects them. If tracking only establishes appointment-link clicks from a defined set of sessions, label that narrower result. Do not silently substitute all appointments as the numerator.
Keep citation reporting in its lane
Bing’s AI Performance report measures how often supported AI experiences cite pages; citation counts do not measure page ranking or placement. This is useful evidence of source use, with a different purpose from website analytics or scheduling records. Bing’s metric definitions make those boundaries explicit.
A rise in citations with flat measured visits is therefore not automatically contradictory. Investigate what pages were cited, which questions they served and whether the reported periods and platform coverage match. Avoid inventing a hidden booking benefit to explain the gap.
Document the attribution method before the claim
Have qualified analytics and privacy reviewers decide which events can appropriately be measured and connected. Keep general marketing reports free of unnecessary patient-level details. A spreadsheet containing personal information is not a prerequisite for an honest business review.
Document attribution windows, missing data, duplicate handling and any changes in the measurement setup. Add operational context such as office capacity, other campaigns or a scheduling interruption. Those factors may affect appointments even when discovery observations remain unchanged.
Self-reported discovery can add context, but “I saw you online” does not identify a particular AI platform. Preserve that ambiguity in the report rather than forcing every response into a marketing channel.
Frequently asked questions
What if our scheduler cannot report confirmation back to analytics?
Measure the steps you can verify and state the gap. Do not relabel a button click as an appointment to make the dashboard look complete.
Can we compare this month with last month after changing tracking?
Only with an explicit explanation of the break. Keep comparable segments where possible and establish a new baseline for newly defined events.