What Is AI Visibility?

AI visibility is a measure of how often AI tools such as ChatGPT, Perplexity and Google's AI features name or link to your practice when patients ask questions in your area. For a health practice, it answers the plain question of whether the new patient who asked an AI for a provider heard your name. It is measured across many questions and many runs, because a single answer tells you little.

How AI visibility works

Traditional search gave you a position, and tools could report it. You were number three for "chiropractor near me," and you could look it up. AI answers do not work that way. A tool writes a fresh answer each time, and it may name two providers or six. So visibility is a rate. Out of a set of runs, in what share were you named or linked? That is a different question from where you rank, and a practice can do well on one and poorly on the other. A clinic that ranks on page one of Google can be missing from an AI answer, because the answer is a written summary and not a list of ten results.

Three things make that number tricky.

• Answers vary. The same question can name your clinic on Monday and skip it on Tuesday. That is normal behavior for these tools, so one run shows little.
• Wording matters. "Best functional medicine doctor near me" and "who treats Hashimoto's in Denver" can produce different lists. A fair score covers the different ways patients ask.
• Personalization adds noise. If you are logged into ChatGPT with saved memory and your own history, the answer may lean toward what the tool knows about you. OpenAI describes Temporary Chats as starting without memory by default, so use one and confirm personalization is off before you record anything.

A good visibility measure has four parts.

• A fixed list of patient questions that covers your services and your cities.
• The same tools each time, such as ChatGPT, Perplexity and Google AI Mode.
• Several runs of each question.
• A record of three things per run: whether you were named, whether your site was linked, and who else appeared.

Divide the runs that named you by the total runs. That percentage is your visibility score.

The same record feeds AI share of voice, your share of all names. It also feeds citation share, your share of linked sources.

What moves the score? The vendors do not publish their rules. Research offers clues, not recipes. An Ahrefs study of 75,000 brands found that YouTube mentions, about 0.737 on a Spearman correlation, and branded web mentions, 0.66 to 0.71, were the strongest correlates of AI visibility, while backlinks were weakly correlated. Ahrefs says correlation is not causation. The practical reading is that being discussed by name in many places goes along with being named, which connects visibility to your AI brand mentions.

The routine for collecting this data over time is covered under prompt tracking.

A score also has more than one dimension. Being present is only the first. Check whether the details said about you are right, because a practice can be named often and described wrongly (see AI brand accuracy). It helps to sort your questions into two groups. Questions that name your practice show whether tools know you. Questions that only describe a need, such as a symptom plus a city, show whether tools would ever pick you. They call for different fixes. A third, slower signal is searches that include your practice name in Search Console. In my judgment, a rise there can mean people heard about you somewhere, including in an answer, but Search Console cannot tell you where.

Example: AI visibility at a health practice

Picture a hormone clinic that tests the question "best bioidentical hormone doctor near me" once, sees its name and stops. This example is illustrative, not a transcript. The next ten runs might name it twice. The honest number is two in ten, or 20 percent, and the trend over months matters more than any single day.

Now widen it. The clinic builds a list of 25 questions across two services and two cities and runs each one three times in ChatGPT. That is 75 runs. It is named in 12 of them.

• Hormone therapy questions, 12 of 45 runs.
• Weight loss questions, 0 of 30 runs.
• Overall, 12 of 75 runs, a 16 percent score.

The overall number is useful, but the split is the finding. The clinic is visible for one service and absent for the other. That tells the owner where to spend writing time.

For real data on how low the starting point can be, here is the Health Biz Scale test. We checked 764 established US health practices across 228 cities between May 11 and Aug 20, 2026, with one check per practice and test runs excluded. We asked GPT-4o-mini, through OpenAI's API and answering from memory with no web search, 2,292 patient-style questions. Only 6 of the 764 practices (0.8%) were named. About 7 in 10 questions named a city, so the model was answering local questions and still left out nearly every practice.

The ChatGPT app searches the web live and can answer differently, so that figure describes memory-only answers, not what patients see in the app. The lesson holds anyway. Established practices should not assume they appear, and a score is the only way to find out.

How to check yours in 60 seconds

1. Open a ChatGPT temporary chat.
2. Ask one patient question, such as "Who are the best functional medicine doctors in Denver for thyroid problems?"
3. Run it three times, opening a new temporary chat each time.
4. Count how many of the three answers name your practice.

Three runs is a rough sample, and it is a starting point only. Treat zero of three as a reason to build a fuller test, and three of three as a reason to try more questions.

What to do about it

1. Write a fixed list of 20 to 30 patient questions covering your services and cities.
2. Run each one several times on the same tools at a regular interval, such as weekly.
3. Record, for every run, whether you were named, whether your site was cited, and who else appeared.
4. Calculate the percentage of runs that named you. That is your visibility score, and you can compute it per service as well.
5. Track the trend over months, not days. In my judgment, expect weeks between a fix and any change in answers, and do not undo a change after one quiet run.
6. Run the list in more than one tool and keep the scores separate. A strong result in one can hide a weak one in another.
7. Use the gaps, such as questions where competitors appear and you do not, to choose what to write next.
8. Check any error you find. A tool that invents a detail about you is the problem called hallucination, and the fix is clearer public facts.