How to fix wrong information about your company in ChatGPT

How do you fix wrong information about your company in AI answers?

Save the exact answer, identify the incorrect statement, and check it against current product evidence. Correct the relevant page you control or ask the source publisher to correct theirs. Make the corrected information accessible to search crawlers, then repeat the same buyer question and record what changes. A published correction improves the available evidence; it does not guarantee a particular AI answer.

The work starts with one claim. “ChatGPT describes us badly” is difficult to investigate. “ChatGPT says our product lacks an integration that our documentation supports” gives your team something it can verify and act on.

What should you record before changing anything?

Keep the full response, not just a screenshot of the offending sentence. Record the question, date, product or model shown, and whether web search was enabled. Save the cited URLs and the exact wording on any source page that appears relevant. Note the region and any conversation history that could affect the answer.

A new chat helps separate the buying question from your earlier instructions. It does not make every result identical or remove all personalization. Use the same documented conditions when you check again.

Separate three observations: whether the company appeared, whether the factual claim was correct, and which product the answer recommended. A company can appear prominently and still be described incorrectly.

How do you decide whether a claim is wrong?

Compare the statement with evidence that applies to the same product, plan, and date. A feature available on Enterprise does not prove that a statement about the free plan is wrong. A current pricing page may contradict today’s answer while supporting what an older article said at publication.

Answer statement	Evidence to check	Possible conclusion
A plan costs a particular amount	Pricing and billing terms for that plan and region	Accurate, outdated, or incorrect
A product lacks an integration	Current integration documentation and plan availability	Supported, contradicted, or unclear
A company meets a security requirement	Approved trust documentation with its scope and date	Supported or not established
A competitor is easier to use	The stated use case and any cited comparison	Opinion or preference, rather than a false fact

Keep unresolved claims unresolved. Ask the product owner when public documentation does not settle the question. Do not publish private contracts or security material merely to give a crawler more detail.

How do you find where the incorrect statement came from?

Open the citation attached to the statement and look for the wording it supposedly supports. A domain in a source list is not proof that it supplied a particular claim. If there is no reliable connection, record the source as unknown.

ChatGPT’s source panel can include both cited sources and other relevant links. OpenAI also warns that search results and citations can be incomplete or incorrect. Asking for sources after the original answer produces a new response, not a record of the original retrieval. See the source-tracing guide for the full process and OpenAI’s search documentation for its source controls.

What should you change on your own website?

Update the page a buyer would reasonably use to verify the fact. Put pricing on the pricing page, integration requirements in the integration documentation, and deployment limits beside the deployment description. State the product name and relevant conditions in visible text.

For an illustrative integration correction, the page might say: “The Salesforce integration is available on Pro and Enterprise. It supports one-way contact sync.” That is easier to verify than a general claim that the product “connects to everything.” Use only details your team can support.

Link to that page from the relevant product or comparison page. Retain a dated change note when the historical distinction matters. Avoid creating several nearly identical correction posts about the same fact.

What if a competitor or another publisher owns the source?

Preserve the page and its relevant wording before contacting the publisher. Send a narrow correction request: the quoted claim, the current evidence, the scope of the discrepancy, and the replacement wording you propose. A publisher may disagree or decline, so maintain an accurate first-party reference page as well.

A competitor’s negative opinion does not become false because it is inconvenient. Check factual statements separately from judgments. Our competitor claim monitoring guide explains how to keep that distinction throughout a review.

What can you send in a factual correction request?

Use a short note that lets the publisher check the discrepancy. Replace the bracketed fields with the evidence you captured:

Your page at [page URL] says “[exact quoted claim].” As of [effective date], [current fact, including plan or regional limits]. The supporting documentation is [evidence URL]. Could you update that sentence to “[proposed wording]”? Please let us know if your comparison refers to a different version or scope.

Keep the request about the specific fact. If the original wording was accurate at publication, say when the fact changed. Preserve the reply and any revision beside the original page capture.

Can you report an incorrect AI answer directly?

Use the provider’s feedback or reporting controls when available and include the exact statement and evidence. Treat the report as a request for review. It does not establish that the provider will change future answers about your company.

Correcting a source page and reporting an answer address different parts of the problem. Keep a record of both actions so later reviewers know what changed.

How do you check whether the correction is accessible?

For ChatGPT search, check that your site permits OAI-SearchBot and that your host allows its requests. Google indexing alone does not establish access for another provider. OpenAI describes crawler access as a requirement for eligibility, with no placement guarantee. ChatGPT search guidance

For Google, inspect the corrected URL in Search Console. Check the indexed version and use the live test to investigate current access problems. Confirm that the page is not blocked or marked noindex by mistake. A successful live test establishes eligibility, not inclusion in search results. Google’s URL Inspection documentation

Google does not require special AI markup for AI Overviews or AI Mode. It points site owners to ordinary search requirements and useful, accessible content. Use question headings because they help readers locate answers, rather than treating FAQ formatting as a ranking guarantee. Google’s AI search guidance

How long does it take for AI answers to change?

There is no dependable provider-by-provider deadline. The result depends on which sources the particular response retrieves, whether the page has been recrawled, and how the answer is generated. Google says recrawling can take days to weeks and a request does not guarantee inclusion. Recrawl guidance

Choose a review schedule appropriate to the claim’s importance. Keep the question and conditions stable. Record the corrected page’s publication date and each later answer, including answers that still repeat the error. A change in one response is an observation, not proof that the edit caused it.

How does KnitKnot help manage the correction?

KnitKnot connects competitive benchmarks with claim intelligence. Your team can inspect the statements behind a recommendation, check disputed claims against the Fact Ledger, and preserve reliable source evidence. Recurring findings stay linked to the response and later measurement.

Start with a claim your team can verify. Use the source-tracing guide to investigate it, then keep the approved fact and correction on the same record.

If you need to check whether the same claim appears across buyer questions, request a KnitKnot scan. It looks for negative claims in AI answers, the sources shown with those answers, and the comparisons where the claims appear. Your team still decides whether a claim is false and what response is appropriate.