Short answer
Correct the sources, not the chat. Assistants rebuild their answer from your website, your Google Business Profile, directories and payer listings, so a correction typed into a conversation changes nothing. Fix every source so they agree exactly, then re-check in two to four weeks and write down what changed.
Key takeaways
- Telling an assistant it is wrong does not update anything for the next person who asks.
- Two thirds of patients who used AI to research a provider met incorrect information, and 60% did not verify it (rater8, April 2026, vendor survey).
- Fix in order: your own site, Google Business Profile, then directories and payer listings, because the later ones copy the earlier ones.
- Some directories overwrite manual edits from upstream data. Find out which ones before you assume a fix held.
The first time a practice sees this, the instinct is to argue with the chat window. It is a reasonable instinct and it accomplishes nothing: the correction lives in that one conversation and dies with it. The answer the next patient gets is rebuilt from the same sources that produced the wrong one.
Why the wrong answer keeps coming back
An assistant composes from what it can retrieve and what it absorbed in training. Retrieved facts come from live pages: your site, your Business Profile, directories, payer listings. Absorbed facts come from a much older snapshot of the web. Only the first kind changes quickly, and only when the underlying pages change.
Fix the sources in this order
- Your own website first. Contact page, footer, every location page, and the structured data in the page source. This is the one source you fully control, and other listings are checked against it.
- Google Business Profile next: name, address including suite, phone, hours including holidays, categories and services. Many downstream listings copy from here.
- The directories your specialty actually uses, plus your state licence board entry, with identical spelling and punctuation. Not approximately identical: identical.
- Payer and insurance directories, which patients trust and practices forget. These are often the source of a wrong phone number that nobody can trace.
- Anywhere you are quoted or listed by a third party: associations, local press, podcast show notes, conference bios.
| What you changed | Typically visible in | How to verify |
|---|---|---|
| Your own site and its structured data | Days to weeks, after re-crawl | Search a distinctive sentence from the page in Bing and Google |
| Google Business Profile | Days to weeks | Check the profile as a logged-out visitor, not from the dashboard |
| Directory listings | Weeks | Re-open each listing yourself, and again a month later |
| How an assistant describes you | Weeks to months | Ask the same fixed questions, logged out, on a monthly cadence |
When the wrong fact is about a person, not the practice
Credentials, training and role are usually absorbed rather than retrieved, so they move more slowly. The lever is the same: make every independent profile agree. Licence board, association directories, LinkedIn, conference bios and your own team page should carry one spelling of the name and one list of credentials.
What if it is defamatory rather than merely wrong?
Wrong hours are a data problem. An assistant stating something damaging and untrue about a named clinician is not, and it is worth documenting immediately: the exact question asked, the full answer, the date, and the sources it cited. Each major assistant has a feedback or reporting path, and that record is what any later conversation, legal or otherwise, will rest on. Screenshots age badly, so save the text.
The realistic expectation
Retrieved details can correct within weeks. Descriptions that come from training data can persist for a long time regardless of what you do, and no vendor offers a way to force an update. That asymmetry is the argument for getting the listings right before a problem exists rather than after.
Questions people ask
Can I contact OpenAI or Google to correct my business information?
There are feedback paths in each product, and they are worth using for damaging errors, but none of them works like a database edit. The reliable route is to correct the public sources the assistant reads, then verify on a fixed set of questions weeks later.
Why does the assistant still show my old phone number?
Because something still publishes it. Usually an unclaimed directory listing, a payer directory, or an old page on your own site that nobody links to any more. Search the number itself in Bing and Google and fix every page that comes back.
How long until a correction shows up in answers?
Weeks for details that come from live pages, once those pages are re-crawled. Months, or not at all, for characterisations that came from training data. Re-check monthly rather than daily, and keep the wording of your questions identical.
Does correcting one assistant fix the others?
No, and that is why source fixes beat product feedback. Each assistant crawls with its own agent and retrieves independently. Correcting the shared sources is the only change that reaches all of them.
Should I keep a record of what was wrong?
Yes. Save the question, the full answer text, the date and the cited sources in one document. It tells you whether corrections are landing, and if an error is ever damaging enough to escalate, it is the only evidence you will have.
Sources
- rater8, 2026 Patient Choice Report (April 2026, n=992 US adults, vendor survey): 66% of the 465 respondents who used AI to research providers met incorrect information; 60% did not verify it.
Published . Last reviewed .
Want this in your business?
Book a call and we will scope exactly which part of your operation an agent should take first.
Book a call with the team