At a glance
A mention rate describes the answers you collected. Read those answers before treating the number as evidence of a recommendation.
Start with a patient’s question
Choose questions about the services and locations you want to understand. A question that already names your office tests whether an assistant can describe it. A question that asks for local options tests a different thing: whether the office appears without being named first.
Keep those two sets separate. Save the wording, location context, engine, model when available, and time of each measurement.
Read the answer behind the number
An office can be named as a recommendation, a comparison, or a passing reference. Those are different contexts even when each contains the name. Look at the passage and the cited sources.
Check the local brand as well as the parent group. Patients may know an office by a name that does not include the DSO. A group-level result can hide that distinction.
Keep the sample size in view
Save the wording, location context, and engine settings for the comparison.
Check the mention, its context, and cited sources. Record failed requests separately.
Use comparable settings and report coverage alongside the mention rate.
An answer can mention an office without recommending it. Keep the original answers with the result.
A mention rate is the share of returned answers that name the target. A failed request or empty response is missing coverage, not an answer that omitted the office.
Keep questions and engine settings consistent when comparing runs. A different question, model, or location context changes the measurement. Grounded API answers also do not reproduce every experience a person may have in a consumer app.
Use the sources to choose the next step
Review the pages assistants cite and compare their factual coverage with your location pages. Look for specific gaps you can verify, such as an unclear service description or outdated office details.
Correct the underlying information and measure again. One change followed by one new mention does not establish cause and effect, and a mention is not a booked appointment.