An AI visibility score is one number that summarizes how often AI assistants name your practice when they're asked a fixed set of patient questions. It's useful for spotting direction over time and nearly useless as a grade on its own, because every tool builds it differently. About 210 people a month search "ai visibility score" and 90 search "ai share of voice" (Google Ads Keyword Planner data via DataForSEO, US, 12-month average to Sep 2026); this page explains what sits behind both numbers.
In short: An AI visibility score condenses a batch of AI answers into one number, usually built from how often you're named, where you appear and which pages are cited. Each tool calculates it its own way, answers vary between runs, and small prompt sets swing hard. Read the trend from a fixed prompt set, not one week's score.
What an AI visibility score measures
Every score starts with a prompt set: the questions a tool asks ChatGPT, Gemini, Claude, Perplexity or Google AI Overviews on your behalf. "Best pediatrician in Raleigh", "who does laser hair removal in Scottsdale", "Spanish-speaking dentist in San Antonio". The score describes how you fared in those answers and nothing else.
That has two consequences. First, the score is only as good as the prompts. A set of easy, branded questions produces a flattering number; a set of competitive local questions produces an honest one. Second, a score measures exposure in sampled answers, not patients. It can't tell you how many people asked those questions or clicked through. For that you need visit data, covered in our guide to traffic from ChatGPT and other assistants.
Common ways tools calculate it
Most scores combine some of these ingredients:
| Method | How it's calculated | What it's good for | Watch out for |
|---|---|---|---|
| Mention rate | Prompts where you're named ÷ prompts run | Simple and easy to audit | Treats sixth place the same as first |
| Position-weighted | Mentions count more when you're named earlier | Rewards being the first suggestion | Weights are the vendor's choice |
| Citation rate | Answers that link or cite your website ÷ answers | Shows whether your own pages are a source | Some answers cite nothing at all |
| Sentiment | Whether the description of you is positive, neutral or negative | Flags inaccurate or negative summaries | Automated sentiment often misreads |
| Composite 0–100 | A blend of the above, scaled | One number for a dashboard | Not comparable between tools |
Hypothetical arithmetic: a practice runs 15 prompts and is named in 3 of the answers. Its mention rate is 3 ÷ 15 = 20%. The same arithmetic appears in the sample Monday brief on the medfy.ai home page, which reads "ChatGPT named you in 3 of 15 answers — up from 1". That brief is a sample for a fictional practice, not real data. Going from 1 to 3 mentions moves the rate from about 6.7% to 20%, which shows how much a small prompt set magnifies each answer.
When a tool shows a 0–100 score without explaining the formula, ask for it. If the vendor can't show how a mention in one answer changes the number, you can't tell real movement from noise.
Why scores move week to week
Assistants write each answer fresh, so the same question asked twice can produce different names. Google states it plainly for its own products: AI Mode and AI Overviews "may use different models and techniques, so the set of responses and links they show will vary," and AI Overviews "often don't trigger" at all (Google Search Central (opens in a new tab)).
Common reasons for a swing:
- Sampling. With 15 prompts, each mention is worth almost 7 points of mention rate.
- Assistant updates. Model changes and new retrieval sources shift answers without warning.
- Changes around you. A competitor's new reviews, a news story or an updated directory profile can change who gets named.
- Changes on your site. A robots.txt rule that blocks a search crawler can remove you from that assistant's search answers. OpenAI says sites that opt out of OAI-SearchBot "will not be shown in ChatGPT search answers," and that robots.txt changes take about 24 hours to register (OpenAI (opens in a new tab)).
- Tool changes. A vendor that edits its prompt set or test location breaks the trend line.
Reading your score: what's meaningful
Treat the score as a smoke alarm, then look at the answers behind it.
- Compare with yourself. Same prompts, same tool, same day of the week. A four-week average smooths sampling noise.
- Find the prompts that changed. A drop driven by one prompt is a different problem from a drop across all of them.
- Read who was named instead. Two weeks of the same competitor replacing you is a signal; random churn isn't.
- Check accuracy. Being named with the wrong hours or an insurance plan you don't take is worse than not being named.
- Mark events on the timeline. A website relaunch, a new listing or a robots.txt change explains movement that the score alone can't.
There is no published "good" score for a practice, and we don't invent one. Medfy's free marketing audit scores AI visibility from 0 to 100 as one of six channels, alongside website, SEO, Google listing, social and ads; it asks Claude and Gemini, with live web search, whether they name the practice and checks that the crawlers behind ChatGPT, Perplexity, Claude and Gemini can read the site. Medfy's weekly tracking then shows whether you're named and who is recommended instead, so you see the answers behind the number. To test prompts by hand first, use the free AI visibility checker.
Improving the inputs, not the number
The fastest way to raise a score is to swap hard prompts for easy ones. Resist it. The number improves and nothing else does.
Work on what assistants read instead:
- Your own pages. A clear page per service that names your city, providers and insurance gives an assistant something specific to quote.
- Listings. Hours, address and services that match everywhere, starting with your Google Business Profile.
- Third-party sources. Directories and local coverage that assistants cite. Our guide to AI visibility for practices covers which sources matter and why.
- Crawler access. Make sure the search crawlers of the assistants you care about aren't blocked.
None of this makes a mention certain. AI answers vary by phrasing, location and day, and no one can promise a practice will be named or recommended. What you can control is whether the information is accurate, findable and readable, and whether you're measuring the result honestly. For the tools that do the measuring, see our overview of AI visibility tools.
Frequently asked questions
What is a good AI visibility score?
There is no reliable benchmark. Every tool calculates its score differently, prompt sets differ, and answers vary by phrasing, location and day, so a number that looks good in one tool means little in another. The useful comparison is with yourself: the same prompts in the same tool, week after week. A score that rises steadily over two months says more than any single reading.
How is AI share of voice calculated?
Count every practice named across a fixed set of prompts, then divide your mentions by that total. If assistants name 24 practices across 15 answers and you appear 3 times, your share of voice is 12.5%. Some tools weight mentions by position or count only the first few names, so check the method before comparing numbers from different tools.
Why did my AI visibility score drop?
Usually for one of three reasons. Sampling: with a small prompt set, one lost mention moves the score several points. Upstream change: an assistant updated its model or sources, or a competitor gained new coverage. Or something on your side broke, such as a robots.txt rule blocking a search crawler or a listing with wrong details. Read the actual answers before reacting.
Is an AI visibility score the same across tools?
No. There is no standard definition. One tool reports the share of prompts where you're named; another weights by position, adds citations or sentiment, or scales everything to 0–100 with its own formula. Prompt sets and the assistants covered also differ. Compare a tool only with itself over time, and ask any vendor to show exactly how its score is built.