There is a category of lost patient that never appears in any report. They asked an assistant for a recommendation, received three practice names, and chose one. Your analytics shows nothing, because there was no impression, no click and no session. The only way to see this traffic is to go and look at it directly, which almost no practice does.

The good news is that checking costs nothing and takes an afternoon, and the resulting list of fixes is usually short and concrete. Here is the audit I run, in a form any practice manager can execute without technical help.

The Audit, Step by Step

What Good and Bad Look Like

The second and third items on that list are the ones practices find most alarming, and they are also the most fixable. Being described inaccurately is usually a data problem in a source you can correct, not an algorithmic judgement about your quality.

Turning the Audit Into a Fix List

Once you have the picture, the work sorts itself into a familiar order, and it is largely the same work that makes a practice visible in ordinary local search, which is why this is efficient rather than an additional burden. Complete and correct your Google Business Profile first, because it is cited constantly. Then reviews, both because they are a direct signal and because review platforms are frequently the source AI systems quote. Then the directory and listing corrections your source-tracing surfaced. Then the site itself: schema, treatment pages, and clinician profiles, so that when a system does reach your website it finds a practice it can describe precisely.

What you should not do is treat this as a separate AI project with its own budget and tactics. The systems reward accuracy, consistency, verifiability and genuine reputation. Those are the same things that have always made a practice findable, now being read by a different kind of reader.

The Prompts to Run, and Why These Ones

The audit only works if the prompts resemble what patients actually type, so it is worth being deliberate rather than improvising. Patients rarely ask neutral questions. They ask from a situation, with a constraint, and often with an anxiety attached, and those constraints are exactly what makes an assistant name one practice over another.

Run each of these in ChatGPT, Gemini and Perplexity separately, because they draw on different sources and produce genuinely different answers. Substitute your own town and treatments, and keep the wording natural rather than keyword-like.

That last prompt is deliberately different from the rest. The first seven test whether you are discovered by someone who does not know you. The last tests what the systems say about you when your name is already known, which surfaces inaccuracies you would never otherwise find: wrong opening hours, a former location, treatments you no longer offer, or a confident summary drawn from a stale directory entry.

Reading the Answers Properly

The instinct is to check only whether you appear, but the answers carry more information than that. Four patterns are worth recording each time, because they point at different fixes.

Turning the Findings Into a Ranked Fix List

Every audit I have run produces findings that sort into the same four buckets, in roughly the same order of return, so you can plan the work before you even see your results.

First, correction work: wrong or missing information in the sources being cited. This is the fastest win available because it requires no strategy, only accuracy, and a single corrected directory listing can change how you are described across multiple assistants. Second, profile work: completing categories and treatments on your Google Business Profile and keeping it active, since it is cited constantly and updates quickly. Third, reputation work: building a genuine ongoing review flow, which moves slowly but compounds and influences both the assistant and the patient reading the answer. Fourth, site work: schema, treatment pages and clinician profiles, which matters most for the deeper questions patients ask once they have your name.

Notice that the site itself comes fourth. That ordering surprises practices who have been sold website packages, and it reflects what the audit actually shows: for the discovery question, the sources that decide the answer usually sit outside your website. The website’s job comes immediately afterwards, when a patient who heard your name goes to check whether the practice looks right for them, which is when treatment pages and real clinician profiles earn their keep.

Making It a Habit Rather Than a Project

The single biggest determinant of whether this work pays off is whether it is repeated. AI answers shift as sources update, as competitors improve their listings, and as the systems themselves change how they retrieve and weight information. A practice that checks once has a snapshot from a moment that has already passed.

The workable version is small and regular: the same prompt set, once a month, recorded in a simple sheet with the date, which practices were named, which sources were cited and whether anything about you was described wrongly. Fifteen minutes, delegated to whoever manages the practice’s marketing. Over a few months that sheet becomes the only genuine record you have of a discovery channel that reports nothing to your analytics, and it will show you your fixes working before any other metric does.