Define what you are checking
Start with a specific decision people make. Instead of asking 'Is my practice visible in AI?' define a small set of realistic questions: 'Which veterinarians near this neighborhood are accepting new clients?' or 'Which dental offices offer implants in this city?' Questions about your own brand are different from open-ended recommendations.
Write down the intended city, service, and audience for each prompt. Avoid adding your practice name to every query, because that turns a discovery test into a brand lookup.
Use several prompts, not one lucky answer
Build a small, repeatable set representing categories such as practice recommendation, specific service, new-patient availability, provider identity, and opening hours. Don't change a question mid-test just because you dislike the answer. Record the exact wording used.
Asking a model variations of the same question dozens of times is not necessarily independent evidence. It may tell you more about the testing setup than about local discovery. Define the sample in advance.
Record enough context to interpret it
For each observation, save the date, AI platform, selected location context where available, exact prompt, response text or a precise summary, and any cited sources. If a platform offers no verified local context, note that limitation. If citations are present, check the cited pages rather than assuming the model's summary faithfully reflects them.
The result can be 'mentioned,' 'not mentioned,' 'information incorrect,' 'question not answered,' or 'not verifiable.' These are more honest labels than a synthetic placement score made from one response.
Separate model behavior from website issues
A missing mention doesn't prove a crawler is blocked. Conversely, a correct answer doesn't prove the site's structured data is accurate. Systems can use search indexes, listings and many kinds of external information. Google's AI search guidance and Bing's webmaster documentation both emphasize traditional indexing and web discovery foundations.
Use the spot check as one evidence stream. Then investigate your public pages for technical accessibility, contradictory facts, weak service descriptions or outdated provider information. Fix confirmed issues whether or not one particular assistant mentioned your practice.
Do not promise a stable position
A snapshot tells you what happened during selected tests. It cannot establish that your practice will always appear, that every user sees the same answer, or that a change to a page will cause an immediate improvement.
The right question after a repair is whether the site's underlying information is now accurate and crawlable. Subsequent AI sampling can provide additional observations, but it should not be sold as causal proof without a stronger experimental design.
How to present findings to a practice owner
A responsible report can include a short table with columns for question, platform, date, observation, and relevant source. Follow it with a paragraph that explains the biggest uncertainties. If a practice wasn't mentioned, say so. If the answer confidently quoted the wrong hours, show the exact discrepancy and investigate the sources.
That is a much stronger deliverable than screenshots chosen only to make the service look impressive. Credibility comes from treating inconvenient results with the same care as favorable ones.
Primary references
Turn findings into useful action.
FindMyPractice.ai identifies technical, content and business-information issues affecting AI search readiness, with specific recommendations for your practice.
Explore audit packages