What Are Branded vs Unbranded Questions?

A branded question is an AI prompt that includes the name of your practice, such as "tell me about [practice] in [city]." An unbranded question describes only the need and the place, such as "best functional medicine doctor in [city]." For a health practice the two are different tests, and treating them as one hides which problem you have.

How branded vs unbranded questions work

Patients ask AI tools in two broad ways, and the tool does different work for each.

When a patient asks a branded question, they already have your name. Maybe a friend sent them, or they heard you on a podcast, or they saw an ad. The AI tool now has to find out who you are. It looks at what exists about that name, such as your website, your Google listing, directories, provider profiles and reviews. The patient is checking you out, and the answer either confirms the decision or creates doubt. Common versions are "is [practice] legit," "[practice] reviews," "does [practice] take insurance" and "what does [practice] cost." These invite the tool to look at reviews and price pages, so what is public in those places can shape the reply.

When a patient asks an unbranded question, they have a need and no name yet. The tool has to decide which practices to mention at all. It leans on the sources it finds for that topic and place, and on which practices those sources describe. This is where new patients are choosing. A practice that is absent here depends entirely on people who already know its name.

Here is the short version.

• Branded questions ask "Are you described correctly?" The test is accuracy. See AI brand accuracy.
• Unbranded questions ask "Are you in the running?" The test is whether you are named, and next to whom.
• Problem questions are a third group worth adding to the unbranded side. These describe a situation and not a service, such as "I am tired all the time and my labs are normal." Many patients start there.

Why keep them apart? Because the causes differ. A wrong branded answer usually traces to a stale directory entry, an unclaimed profile or a page on your own site that disagrees with another. A missing unbranded answer usually traces to the sources the tool trusts for that topic not mentioning you, or your pages not answering the question directly.

On the unbranded side, outside sources appear to carry a lot of weight. The MGMA article does not say whether its prompts named a practice, but its example reads as a provider search, so I treat it as unbranded evidence. An MGMA analysis of 33,340 ChatGPT citation events across 200 practices, 20 metros and 10 specialties (May 2026, written by the founder of a commercial service in this space) found that hospital staff rosters and specialty association credentialing pages together supplied 49.6% of the cited surface. Practice-owned websites were about 23%. A practice can be strong on its own site and still be absent where the tool looks first.

Because AI answers change from run to run, one result tells you little. Ask each question several times and read the pattern. See AI answer variability for why.

To keep a steady record, use the method in prompt tracking. The same method also shows where you stand against other practices.

Example: branded vs unbranded questions at a health practice

This example is illustrative, not a transcript. A functional medicine clinic in a mid-sized city tests six questions in one AI tool.

The three branded questions are "tell me about [clinic] in [city]," "does [clinic] take insurance" and "who are the providers at [clinic]." The answers name the clinic and the city correctly. But one answer lists a physician who left the practice two years ago, and another says the clinic accepts insurance when it is cash-pay. The clinic is visible and partly wrong.

The three unbranded questions are "best functional medicine doctor in [city]," "who treats Hashimoto's near [city]" and "I am exhausted and my doctor says my labs are fine, who can help in [city]." The answers name a hospital system, two other clinics and a directory page. The clinic does not appear.

A blended score would average these into something that looks middling. The two columns tell a clearer story.

The branded problem is accuracy. The fix is to find where the old provider and the insurance claim come from, which is often a directory or an old page, and correct that source. The unbranded problem is presence. The fix starts with the links the tool showed. If a directory page and a hospital roster appear again and again, the clinic checks whether it is listed on them, and whether its own pages answer the patient's exact question in the first sentence.

Neither fix is complicated. The point is that they are different jobs, and the clinic would have spent its time on the wrong one if it had only typed its own name.

How to check yours in 60 seconds

1. Open one AI tool in a fresh chat with search on.
2. Type "tell me about [your practice] in [your city]." Read the answer for your name, specialty, providers and payment model. Mark each line right or wrong.
3. Start a new chat. Type "best [your specialty] in [your city]." Note whether you are named and who else is.
4. Copy every link the tool shows into a sheet, with the date and the tool.
5. Repeat step 3 with one problem-style question, written in the words patients use.

Record the branded and unbranded results in two columns.

What to do about it

1. Write three to five branded questions and three to five unbranded questions. Draw the wording from intake forms, front desk calls and reviews.
2. Keep the wording fixed and run each question several times, in a fresh chat each time.
3. Log the tool, the date, who was named and every link shown.
4. For branded answers, mark each line right, partly right or wrong. Trace each wrong line to the page it came from and correct that page.
5. For unbranded answers, list the outside sites that repeat. Claim or correct your listing on each one that fits.
6. Check that your own service pages answer the patient's question in the first sentence.
7. Report the two columns separately each month and compare to the last month.
8. Use invented patients in every test. Do not paste real patient details into an AI tool.