You cannot reliably attribute appointments to AI answers by looking at traffic alone. For a multi location clinic, the practical way to measure AI assisted discovery is to define a measurement spec that tracks a stable set of pages AI systems tend to cite, captures patient intent events, separates phone intent from connected and qualified calls, and reports lift as a segment comparison against a baseline instead of hunting for a single “AI” source label.
This is for the marketing manager who owns pipeline reporting and needs something defensible across locations, service lines, and scheduling workflows, even when referrer data is missing or messy.
What “worked” looks like for a clinic
AI assisted discovery is not a channel you can count on seeing in one neat bucket. Many AI experiences send traffic that looks like ordinary referrals, direct, or organic, and some send no click at all. So define success in clinic terms, then measure the behaviors that precede it.
For most clinics, the measurable outcomes are:
- Appointment requests (start and submit, for each location and service line).
- Calls (tap to call, connected calls, and qualified calls).
- Location intent (directions, map opens, and similar actions that indicate “I am going there”).
The goal of the spec is consistency, so leadership can trust week to week reporting, and so you can scale from one location to many without changing definitions every month.
Build an AI discovery landing page cohort you can keep stable
Start by deciding which landing pages count as “AI citeable” for your organization. You are not trying to guess which AI system mentioned you. You are defining a cohort of pages that are likely to be used as answers, summaries, or citations because they clearly explain a service, a provider, or a location.
Good cohort candidates for clinics:
- Service line pages that explain a condition, treatment, or specialty and have a clear next action.
- Location pages with address, hours, insurance notes, and the correct phone number for that location.
- Provider profile pages when patients commonly search by clinician name or credentials.
- Scheduling entry pages (the last page on your site before a third party scheduler, if you use one).
Then group those pages into a small number of cohorts you can report on without arguing about edge cases. For multi location clinics, the most useful grouping is usually two dimensions: service line and location. If a page represents both, pick one primary grouping rule and stick to it.
Keep the cohort stable by writing down inclusion rules. Avoid rules that change every time you redesign navigation or rename a service. Use URL patterns and page templates, not menu labels.
Instrument patient intent events, not just pageviews
Once you have the landing page cohort, define the minimum set of events that represent patient intent. These events should be consistent across locations, even if the on page content differs.
For clinics, the intent events that usually matter most are:
- Appointment request start (click from your site into the scheduling flow, or the first step of an on site request).
- Appointment request submit (a completed request or confirmed booking, depending on what you can measure).
- Contact form submit (including “request a call back” if you offer it).
- Location actions like directions and map opens, plus email clicks if that is a real intake path for your clinic.
If scheduling happens on a different domain or vendor hosted system, you may only be able to measure the handoff click reliably. That is still useful, as long as you label it honestly as “start” and do not pretend it is a completed appointment.
Where it fits, add campaign tagging to links you control so you can separate true experiments from background noise. Google’s URL builder supports adding utm parameters to destination URLs to identify which campaigns refer traffic. Use this for email, paid social, and partner links. Do not expect it to solve AI attribution, but it prevents you from miscrediting AI when the lift was actually your own campaign.
Use a call outcome ladder so “calls” mean the same thing everywhere
For healthcare, phone reporting breaks down when all phone activity gets counted as one number. A tap to call is intent, not a conversation. A connected call is a conversation, not necessarily a patient you can schedule. Your measurement spec should define a ladder of outcomes that your team can apply consistently across locations.
At minimum, separate:
- Phone intent (tap to call on the website, click to call from a location page, or similar).
- Connected call (a call that was answered or reached voicemail, depending on how your phone system reports connections).
- Qualified call (a call that fits your intake definition, for example appointment sought, insurance accepted, service line match).
The key is not the perfect definition, it is the shared definition. If one location counts refill requests as qualified and another does not, your “AI is working” story will swing wildly based on staffing and call handling, not demand.
Report lift with segment comparisons, not an “AI” source label
Because AI referrals are inconsistent, your reporting view should focus on whether the AI prone landing page cohort is producing more high intent actions than it used to, compared against a baseline and compared against non cohort pages.
A practical weekly view for a clinic usually includes:
- Cohort segment: sessions that landed on your AI discovery cohort pages.
- Control segment: sessions that landed on other pages (or a stable subset like your homepage and general info pages).
- Conversion events: appointment start, appointment submit when available, contact submit, phone intent, connected calls, qualified calls.
- Breakouts: by location and service line, so one high performing site does not mask another that is slipping.
Use a baseline window that you keep consistent. If you change the cohort definition or event definitions, treat it as a new measurement era and note it, otherwise you will spend every meeting debating whether the numbers are comparable.
AI Assisted Discovery Measurement Spec for Clinics
Fill this in once for one location, then reuse it across the system with location specific phone numbers, scheduling URLs, and outcome definitions.
| Spec Section | Your Clinic Definition (fill in) |
|---|---|
| Landing Page Cohort Rules |
Include page types: (service line pages, location pages, provider pages, scheduling entry pages) URL patterns to include: (example: /locations/…, /services/…, /providers/…) Exclude page types: (blog, news, careers, policies, patient portal, internal search results) Grouping fields: Primary = (location or service line), Secondary = (the other) Cohort change policy: (who approves changes, how often, how you note changes in reporting) |
| Event Taxonomy (High Intent) |
Appointment request start: (what click or step defines “start”, where it happens) Appointment request submit: (what defines “submit”, if measurable) Contact form submit: (which forms count, how spam is handled) Directions or map open: (what counts as the event, where it is tracked) Location page phone intent: (tap to call definition, which numbers, which pages) |
| Call Outcome Ladder |
Phone intent: (website tap to call, other tracked intents) Connected call: (answered and or voicemail, minimum connection rule if any) Qualified call: (your intake definition, include and exclude examples) Dispositions list: (new patient, existing patient, billing, wrong number, other) Location consistency: (how locations apply the same outcomes) |
| Weekly Reporting View Checklist |
Segments shown: (cohort vs control, plus location and service line breakouts) Metrics shown: (events above, plus totals and rates you rely on internally) Baseline window: (your chosen baseline period and comparison method) Annotation log: (site changes, tracking changes, campaign launches, staffing changes) Data quality checks: (missing events, sudden drops, phone number changes, form errors) |
Limits and tradeoffs in AI assisted discovery measurement
Attribution gaps are normal. Some AI experiences do not pass clean referrer data, and some patient decisions happen without a click. Your spec is about measuring lift in the behaviors you can observe, not proving a single appointment came from a single AI answer.
Privacy and consent constraints may limit what you track. Healthcare organizations often have stricter internal rules than other industries. Keep your events focused on intent actions and operational outcomes, avoid collecting sensitive details in analytics events, and align with your compliance team’s guidance.
Cross domain scheduling can hide completions. If booking happens on a vendor system, you may not be able to measure submits without additional integration. Do not “fill in the blank” by assuming starts equal bookings.
Do not treat “AI” as a source of truth. Avoid creating a dashboard tile that claims “AI appointments” as a single number. It will be brittle, and it will encourage over interpretation. Stick to cohort lift and intent events that your clinic controls.
A 60 minute plan to draft the spec and instrument one location
- Minutes 0 to 15: Pick one location and 5 to 10 priority pages, at least one service line page and one location page. Write your cohort inclusion and exclusion rules.
- Minutes 15 to 30: Define your minimum event set for that location, appointment start, contact submit, directions or map open, and tap to call. Write exact definitions in the spec.
- Minutes 30 to 45: Write your call outcome ladder and dispositions list, then confirm the location manager agrees to use it consistently.
- Minutes 45 to 60: Create a simple weekly view in your analytics and call reporting that compares cohort vs control and includes annotations for changes.
If you want a second opinion on the spec and help implementing clean reporting across web events, calls, and appointment requests, see Danbury Web’s Digital Marketing Services.
