What Is Citation Share?

Citation share is the percentage of AI answers, in a fixed test set, that link to your website as a source. For a health practice, it differs from a brand mention. A mention means the AI said your name. A citation means it linked to your page as support for what it said, which is the version that can send a patient to your site.

How citation share works

AI tools can name your practice without linking to you, and they can link to your page without naming you in the answer. Both are useful, but they are different things to measure. Mentions show awareness. Citations show that a page was chosen as a source, which tells you the content is working. A clinic that tracks only one of them sees half the picture, and the half it misses is often the one it can fix.

Citation share turns that into a rate. It is a simple idea with a few traps, so the method matters more than the math. You fix a list of patient questions, run them, and check the sources under each answer. Each answer gets a yes or a no for whether any link points to your domain. Divide the yeses by the total answers.

Here are the pieces you need.

• A fixed question list that covers your services and cities.
• The same tools every time, with each tool scored on its own line.
• Several runs per question, since answers vary.
• A record of which page of yours was linked, not just that a link existed.

Because answers vary run to run, a single screenshot with your link in it is a data point and should not be treated as your score. Run ChatGPT checks in a temporary chat, which OpenAI describes as starting without memory by default, so personalization does not skew results.

What earns a citation? The vendors do not publish their rules, so treat what follows as clues. An SE Ranking analysis covering 129,000 domains and 216,524 pages found that referring domains, meaning other sites that link to a page, were the strongest predictor of ChatGPT citations. Domains with up to 2,500 referring domains averaged 1.6 to 1.8 citations, while those with over 350,000 averaged 8.4. That is correlation from one vendor's data. It suggests that outside recognition goes along with being linked.

Citation share also depends on the sources that exist in your market. If the pages tools lean on are hospital rosters and association directories, your own site competes with those. A practice with a strong site can still have a low share if tools prefer a hospital roster or a directory for the same question. Look at who is cited when you are not. That list is the real competition, and it is covered in more depth under AI citation.

Citation share and AI share of voice are siblings. One counts links to your domain, and the other counts your name against everyone else's. Run both from the same set of answers so the work is not duplicated. In your sheet, give each answer a row, then add one column for names and one for links. A single pass through the answers fills both.

Two refinements make the number more useful. First, separate domain from page. Having your domain cited is not the same as having the right page cited, so log the exact URL and check that it is the service page you would want a patient to land on. Second, report share for each family of questions, such as questions that name your practice and questions that only describe a need. A healthy score on the first family and a zero on the second suggests patients who already know you can find you, while new patients may not.

Add a second column to your sheet for answers where a source the AI cites lists your name, even if that source is not your own site. Gaps in that column are your to-do list for listings and outside pages. A source map built from these rows shows which outside pages shape answers in your market.

Example: citation share at a health practice

Consider a functional medicine clinic that is named often because a local directory lists it, yet its own site is not linked. Its mention rate looks fine and its citation share is zero. That points to the fix. The clinic's pages are not answering questions clearly enough to be used as a source. A different clinic with strong guides but a thin directory presence may show the opposite pattern.

Now the arithmetic, with figures made up to show the method. This example is illustrative, not a transcript. The clinic runs 20 questions twice each in ChatGPT, which is 40 answers, then does the same in Perplexity.

• ChatGPT. 40 answers, 2 linked the clinic's domain. Citation share, 5 percent.
• Perplexity. 40 answers, 10 linked the clinic's domain. Citation share, 25 percent.
• Pooled. 80 answers, 12 linked it, which is 15 percent.

The pooled 15 percent hides the real story. The clinic is cited five times as often in Perplexity as in ChatGPT, and the owner would look at what Perplexity picks up that ChatGPT does not. Scoring each tool separately makes that visible.

Real data shows where citations come from. An MGMA-published study by the founder of a commercial service, covered 33,340 ChatGPT citation events for independent practices, in May 2026 using GPT-5.5 through the API. Practice-owned websites made up about 23% of the cited surface. Hospital staff rosters were 27.5%, specialty association credentialing pages 22.1% and national medical directories 8.3%. Review aggregators were 3.8%. The article reports no direct citations of Google Business Profile. It is one vendor's study of one model, so treat it as a clue. Use it to decide which kinds of pages to check first, and let your own test set tell you what holds in your specialty and city.

How to check yours in 60 seconds

1. Open a ChatGPT temporary chat and ask one patient question twice in separate chats.
2. Open the links or sources shown under each answer.
3. Check whether any link points to your domain.
4. Count the answers with your link out of two. That is a rough starting reading.

If you want a second reading, repeat the steps in Perplexity and keep the two results apart. Note the page that was linked, since a link to your homepage tells you something different from a link to a specific service page.

What to do about it

1. Use the same fixed question list you use for prompt tracking.
2. Run each question several times per tool.
3. For each answer, record yes or no for a link to your domain, plus which page was linked.
4. Divide answers that cite you by total answers for your citation share.
5. Check which of your pages get cited, and write more like them.
6. Look at who is cited when you are not, then compare their pages to yours. If the cited pages are association or board listings, get your listing completed and accurate there.
7. Check that each page you want cited is public, indexed and answers one question in its first two sentences, since a page that is hard to read is hard to use as a source.
8. Earn links from relevant outside sites, since referring domains were the strongest predictor in the SE Ranking data.