TL;DR
Treat a person's reported discovery source, website referral data and appointment records as different evidence. They can inform one another without proving a single causal path. Keep the question neutral and the reporting appropriately limited.
Decide what you actually want to learn
An owner may want to know whether people encounter the practice through AI tools. That is a different question from determining whether a specific article caused a confirmed appointment.
Begin with the narrower question. If the practice already asks how people found it, review whether the options allow an honest answer such as an AI assistant, a search engine, another person, more than one source or uncertainty. Do not force a choice that hides a mixed journey.
Any change to intake or data collection should follow the practice's normal privacy and operations approval process. This is a measurement design discussion, not authorization to collect new information.
Keep three evidence types separate
| Evidence | What it can support | What it cannot establish alone |
|---|---|---|
| Self-reported source | What the person remembers or chooses to report | A complete sequence of every prior touchpoint |
| Website referral or event data | An observed visit or action under the configured system | An appointment outcome outside that system |
| Operational appointment record | The practice's recorded scheduling outcome | Which marketing exposure caused it |
Do not assume a reported discovery source was the final step before contact. Ask what each record actually captures, and preserve uncertainty about the rest of the journey.
Use neutral wording
Avoid asking, “Did our AI visibility work bring you here?” That suggests the desired answer and assumes knowledge the person may not have. A simple optional source question should permit uncertainty and multiple influences where the practice's process supports it.
Do not ask people to share private conversations with an assistant or copy health-related prompts into a marketing form. The business question can usually be explored at a broader level.
Document technical boundaries
If a journey crosses from the practice site to a separate scheduling domain, measurement depends on the systems and approved configuration. Google's cross-domain documentation describes a specific setup for supported properties; it should not be assumed to exist merely because both sites are accessible. Google Analytics cross-domain guidance.
Have the technical and privacy reviewers determine what is available and appropriate. Label unmeasured steps rather than filling them with estimates presented as facts.
Report agreement and disagreement honestly
Report the actual source-question responses and recognizable website referrals in separate series. If their counts differ, first compare their definitions and coverage. They represent different observations and need not agree.
Use a shared measurement dictionary for citations, visits and appointments. Explain the definitions, missing data and reporting period. Do not add unlike counts together into a total of “AI patients.” Use the evidence to decide whether deeper investigation is worthwhile.
Frequently asked questions
Can self-reported discovery prove return on investment?
It can contribute context, but it does not by itself establish causation, revenue or incremental results.
What if the sample is small?
Report the actual count and uncertainty. A small number can generate questions for further review without supporting a broad growth claim.