How we measure AI visibility, and what a report actually proves.

Inside an AI visibility report

Most AI visibility numbers are hard to check. Ours are built so you can open any one of them and see exactly where it came from.

Start with the questions patients ask

We agree a watchlist with you: the questions someone near you would ask about the care you provide, worded the way people speak, such as which practices nearby see children for ADHD or which clinics take a certain insurance plan.

A good watchlist is small and specific. Twenty questions that match your services tell you more than two hundred generic ones.

Ask more than once, and over time

Assistants give different answers to the same question on different days. Each question is checked repeatedly, across the assistants you choose, so a single lucky or unlucky answer does not drive decisions.

Trends over weeks matter more than any one day’s result.

Named and linked are counted separately

An answer can mention your practice by name, link to your website, both or neither. We record those separately, because a mention builds awareness while a link can send a visit you can measure.

Only the names you approve count, so a common word in your practice name never inflates the result.

Every number leads back to an answer

Each counted result keeps the question asked, the assistant, the date, the full reply and the sources it cited. Open any figure in the report and you can read the answers behind it.

If a result looks wrong to you, it can be checked rather than argued about.

What the report cannot tell you

Assistants do not explain why they chose someone, and a sample of questions is never every question. Someone may read an answer and phone you without clicking anything, so the report cannot count patients.

We state those limits in the report itself, next to the numbers they affect.

Where the questions come from

Watchlist questions are drawn from three places: the services and towns your team confirms, the search terms your own Search Console data shows, and what your front desk hears on the phone. Each one is worded the way a patient would ask it, including the plain words people use for a condition rather than only the clinical term.

The list should stay stable long enough to show a trend. Changing it resets the comparison, so changes are best made on purpose and noted with a date. When you add a service or a location, new questions for it start their own history instead of making an earlier month look better or worse.

Why the same question gets different answers

AI assistants write each reply fresh. The wording can change between two runs a few minutes apart, and so can the practices they name. Answers can also vary with a person’s location, account settings and earlier conversation, none of which a monitoring check shares with a real patient.

That is why one result is never treated as a finding. A practice named in most checks of a question over several weeks is in a different position from one named once. When results swing widely, the honest reading is that the assistant has no settled answer to that question yet.

How to read your named and linked results

Start with the questions, not the totals. Which questions name you, which name other practices instead, and which name no local practice at all? Each pattern points to different work: a page to strengthen, a source behind another practice’s mention to understand, or a question no one answers well yet.

Then read the replies that matter most. Being named with the wrong ages, an old address or a service you do not offer still counts as named, but it is a problem to fix. Open those answers first and follow the cited sources back to the page or listing that carries the error.

Read it beside your search data

Answer samples show what assistants said. Google Search Console shows how your pages did in Google Search, and Google includes traffic from its AI Overviews and AI Mode inside the same Web performance numbers rather than reporting it separately. Bing Webmaster Tools has an AI Performance report, in preview, that counts how often your pages are cited in Bing’s AI answers.

Bing says that report shows how often content is cited, not ranking or the part a page played in an answer. None of these sources sees phone calls from people who never clicked. Put them side by side and look for agreement: a question that starts naming you, a page that starts being cited, and more inquiries about that service.

Questions to ask of any AI visibility report

Whoever does the measuring, ask the same things. Which assistants were checked, and which were unavailable? How many questions, how many runs of each, and over what dates? Who decided which names count as your practice? Can you open the reply behind each number? Are other practices measured the same way?

Be careful with a single visibility score or share-of-voice percentage that cannot be traced back to answers. Tools define these numbers differently, so two vendors can report very different figures for the same practice. A number you cannot open is hard to act on and easy to misread.

How often to look, and what to do next

For most practices, a monthly review is enough. Look at the questions whose answers changed, open the replies behind the change and agree on one or two fixes. Day-to-day movement is mostly noise, and reacting to it uses staff time without improving anything.

After a fix is approved and published, give it time. The page has to be recrawled and the assistant has to draw on the new version, and neither happens on a schedule you control. Recheck the same questions over the following weeks before deciding whether the change helped.

Turning a finding into a fix

The useful output is not a score. It is a short list such as “assistants say you do not see children, but you do” or “your insurance list is cited from an outdated directory”, each linked to the page or listing to correct.

Corrections are drafted from facts you confirm, approved by your team, and then rechecked in later samples to see whether the answers changed.