You ask ChatGPT about your own business and get an answer that is plainly wrong. It lists a service you stopped offering, quotes an old price, or puts your company in the wrong category.
Before correcting anything, identify which information sources and tools contributed to the answer. ChatGPT can draw on the model, the current conversation, saved memory, uploaded files, connected sources, and tools such as ChatGPT Search. The visible Search and tool activity is the best place to start, and any citations that appear give you specific pages to inspect.
TL;DR
- First check whether ChatGPT used Search or another tool. Search responses may include citations or a Sources panel, but those elements are not guaranteed.
- If a cited page is wrong, correct that page and the matching facts on your own site.
- If no source supports the claim, save the answer, send thumbs-down feedback, and publish a clear correction on the page a customer would reasonably visit.
- Use thumbs-down feedback for an ordinary incorrect answer. Use formal content reporting for safety, legal, or policy concerns. Neither path is a business-listing editor or a guaranteed correction process.
- There is no reliable propagation timeline. Verify the result in a new chat and keep a record of the mode, prompt, answer, and sources.
Start by identifying the answer type
An answer without web Search may draw on the model, the current conversation, saved memory, uploaded files, or connected sources. It may be stale, it may misunderstand the prompt, or it may state something that no source supports. Record which inputs and tools were active before deciding where the error came from.
A Search answer retrieves web results and may include citations or a Sources panel. The sources can still be incomplete, outdated, or wrong. Review the visible Search activity and inspect every cited page that is available. Missing citations alone do not prove that Search did not run.
Do not try to reverse-engineer the model’s training set from one answer. OpenAI does not provide a source-by-source history for a business claim, and a public page is not proof that it appeared in training data.
| What you see | Likely case | Best next step |
|---|---|---|
| Search activity, citations, or a Sources panel | Web search may have contributed | Open the Search details and every available cited page |
| An uploaded file, connector, or another tool is shown | Tool-grounded answer | Inspect the named source and confirm whether it contains the wrong claim |
| No external tool or source is shown | Model-and-context answer | Reproduce it in a new chat, note whether memory is active, then ask ChatGPT to search and cite sources |
| The answer changes after earlier messages | Conversation context may be steering it | Start a clean chat and use the shortest neutral prompt that reproduces the issue |
| No cited page supports the claim | Generated error | Send thumbs-down feedback and publish a direct, accurate statement on your site |
How to fix what ChatGPT says about your business
Work in this order so you do not spend a week cleaning listings that had nothing to do with the answer.
1. Save evidence before the answer changes
Copy the prompt and full answer. Record the date, whether Search ran, and every citation. Open the cited pages and save the exact sentence that supports the wrong claim. If you cannot reproduce the error in a clean chat, conversation context may have caused it.
2. Find the current public record
Search the exact wrong phrase in quotation marks. Then check the pages a customer would trust for that fact: the relevant page on your site, your Google Business Profile, major industry profiles, and any source ChatGPT cited. Do not assume every directory matters. Start with the sources that appear in the answer or rank for the fact in question.
3. Correct the source, not just the homepage
Put the accurate fact where it belongs. A discontinued service belongs on the service page. A new address belongs on the contact page and location profiles. A pricing correction belongs on the pricing page, if you publish prices.
Write the correction for a person first. A clear heading and a direct sentence usually do more good than a paragraph written for a crawler. If you use structured data, make sure it matches the visible page. Schema can describe a fact, but it cannot override conflicting copy.
If a cited publisher or directory owns the wrong page, use its normal correction process. Keep a record of what you submitted and when. An unreturned email is not proof that the page changed.
4. Send the right kind of feedback
For an ordinary factual error, use the thumbs-down control under the response and explain what is wrong. If current public evidence proves the correction, include it in the feedback when the interface allows.
Use OpenAI’s formal content-reporting process when the response may violate OpenAI’s terms or applicable law, including a serious safety or legal concern. It is not a general business-corrections form. Neither feedback route edits a business listing, and OpenAI does not promise that it will rewrite future answers or respond on a fixed schedule.
5. Verify both cases again
After the source itself is correct, start a new chat. ChatGPT may search automatically even when you do not request it, so record the visible Search and tool activity for the first run. Then run a second version that explicitly asks ChatGPT to search the web and cite sources. The Search answer is the useful near-term check because you can inspect available sources. A different answer without Search does not prove the model’s underlying knowledge was permanently changed.
Does fixing one AI engine fix them all?
No. Each product has its own models, search partners, indexes, ranking systems, and interface. Correcting a public source is still useful because several products may retrieve it, but no change is guaranteed to appear everywhere.
Verify each platform that matters to your customers. Record whether its answer used live search and which sources it cited. Treat an accurate answer in one product as one observation, not proof that the record is fixed across the market.
How to monitor what AI says about your business
Use a small, fixed prompt set once a month. A practical sample can start with five prompts:
- Your brand name alone:
"[Business Name]" - Your brand name plus your core service:
"[Business Name] [primary service]" - Your category plus your location or market:
"[service type] in [city/industry]" - A direct competitor comparison:
"[Business Name] vs [Competitor]" - A question your buyers actually ask:
"who is the best [service type] for [use case]"
For every run, log the platform, exact prompt, answer, whether search ran, and cited URLs. AI answers vary, so flag an issue when it repeats or when the wrong claim is serious enough that one occurrence matters. This is a monitoring sample, not a rank tracker with a stable position.
The log will show whether an error repeats, which sources appear with it, and whether the issue is serious enough to justify more work. It can also support a broader GEO and AI search strategy without pretending that one answer is a stable ranking.
Correct the record, then verify the answer
There is no single switch that updates what ChatGPT says about a company. There is, however, a practical sequence: identify the sources and tools involved, correct the cited or authoritative source, send the appropriate feedback, and test again in a clean chat.
That process will not produce a guaranteed timeline or an identical result in every AI product. It will leave customers with a clearer public record and give search-grounded answers a better source to use.
If you want help auditing what AI engines currently say about your business and building a correction plan, Book a free 30-minute call.
