ChatGPT is the assistant most patients have used, so it is worth knowing what it says about you. It behaves differently from the others in one important way: it may answer from memory or from live browsing, and the two produce very different results.
Two modes, two behaviours
- From training data. Answers drawn from what the model absorbed during training. This favours well-established practices with a long web history, and it can be out of date. A practice that moved recently may still be described at its old address.
- With live retrieval. The assistant searches the web and composes from current sources, usually with citations. Current, and more responsive to work you have just done.
You cannot control which happens. What you can do is make sure both paths lead somewhere accurate: a long-standing consistent record for the memory path, and correct current sources for the retrieval path.
What it draws on
For local care questions, retrieval typically surfaces your website, map and profile data, the significant directories, and review content. Established sites and recognised directories carry more weight than obscure ones, which is another argument for depth on a few good directories rather than presence on many weak ones.
Checking what it says about you
Do this yourself, today. It takes fifteen minutes and it is frequently sobering:
- Ask who provides your treatments in your city, without naming yourself, and note whether you appear.
- Ask directly about your practice by name and read the description carefully.
- Ask for your address, phone, and hours, and check every character.
- Ask which treatments you provide.
- Ask whether you accept a specific insurance plan.
- Ask for alternatives to your practice, which reveals who it considers your competitors.
Use a fresh conversation for each, since context carries over and contaminates the result. Repeat a few days later, because single answers are unreliable.
Reading the results
Sort what you find into three buckets, because they need different responses:
- Absent. You are not named at all. Usually a coverage or resolution problem: thin presence in retrievable sources, or contradictory data.
- Present but wrong. Named with an old address, a dead phone number, or treatments you no longer offer. This is a source problem, so find which source still says it and fix that.
- Present and accurate. Verify it stays that way, and check the justifying sentence describes you as you would wish.
Wrong is more urgent than absent. Being described with a disconnected number actively loses patients, whereas being absent is a missed opportunity.
Fixing what it says
There is no correction form, so you work through the sources:
- Find the source of the error. If retrieval cited pages, read them. The wrong fact is usually sitting on a directory you had forgotten.
- Correct upstream first, as in chapter four, since aggregators and NPI data regenerate errors.
- Make the correct facts unambiguous on your own site, in plain text rather than only in images or PDFs.
- Add structured data so the facts are machine-readable.
- Be patient. Retrieval reflects changes within weeks, training data can take much longer.
Practical expectations
You will not control this engine. You can be accurately and consistently described everywhere it might look, which is the only durable strategy available. Practices that complete chapters two through four generally find themselves named; practices with contradictory data generally do not.
Gemini is next, and it is the most winnable of the engines.