How to fix wrong information about your brand in AI answers

Find where the wrong fact comes from, correct it at every source you control, ask the owners of other sources to correct theirs, and report the answer through each engine's feedback route. Do all of it, because no single step is guaranteed to change the answer on its own.

The makers are candid about this. Google says AI Overviews "can and will make mistakes". OpenAI says ChatGPT's search results and citations can be incomplete, outdated or incorrect. What they document is how answers are built and how to send feedback, and that is what this guide works from.

Step 1: Find where the answer comes from

1. Ask the question in the engine that got it wrong, and save the answer with the date and the exact wording you used.
2. Save the answer's link if the engine offers one. Perplexity asks for the URL of the query in reports of inaccurate answers.
3. Open every source the answer cites and find the sentence that carries the wrong fact.
4. Ask again on another day and in the other engines. As our methodology notes, answers vary between runs, and a single wrong answer may not repeat.

Most answers that cite sources are built from pages the engine retrieved. Google describes its AI features as grounded in pages its ranking systems pull from the Search index, and Anthropic notes that the quality of Claude's answers depends on the sources it references. An answer that cites nothing may instead reflect what the model learned in training. Anthropic warns that Claude may not have been trained on the most up-to-date information.

Step 2: Fix what you control

State the correct fact once, plainly, on the page that should own it: pricing, product, about or documentation. Put it in the HTML the server sends, with the date it was last checked. In our test of nine AI visibility pricing pages, two gave a crawler no readable price. When your own page has no readable figure, any figure an engine repeats has to come from somewhere other than your page.

Then clean up around it:

• Update or redirect old pages that still carry the old fact, such as past pricing pages and launch posts.
• Make sure the engines' search crawlers can read the page, as covered in robots.txt for AI crawlers.
• Add Organization markup with sameAs links to your official profiles, as described in schema markup for AI search.

Finally, tell the engines the page changed. For Google, ask it to recrawl the page; Google says crawling can take anywhere from several days to several months. For Microsoft's engines, Microsoft says IndexNow, which notifies participating search engines when content is added, updated or removed, helps AI systems reference the current version of a page.

Step 3: Fix Google's own records

Google's knowledge panels are generated from information across the web, and Google says they update automatically as that information changes. It also takes corrections directly.

1. Search for your company and find its knowledge panel.
2. Click Claim this knowledge panel and verify through one of the official sites or profiles Google lists.
3. Click Suggest edits, then Edit next to the wrong fact.
4. Describe what is wrong and link to public pages that support the correction.

Google says it reviews feedback from verified users within a few days. Not every panel can be claimed; for those, anyone can use the Feedback link on the panel. Google's guidance on AI features also lists keeping Business Profile and Merchant Center information up to date. Our glossary entry on the Knowledge Graph explains where panels come from.

Step 4: Ask other sites to correct theirs

Third-party pages carry a lot of weight in AI answers. In our study of 3,531 answers from seven engines, roundups alone took 52.5 percent of the citations to the 100 most-cited sites. An out-of-date list can keep being read until someone updates it.

Email the author of each page that repeats the error. Quote the wrong sentence, give the correct fact in one line, and link to the page on your site that shows it. How to get into "best tools" roundups covers how to approach list authors.

If the error is on Wikipedia, follow its conflict of interest guideline. It says editors with a conflict of interest should disclose it, and are strongly discouraged from editing the article directly. Propose the change on the article's talk page instead, with sources.

Step 5: Report the answer to each engine

OpenAI also runs a content reporting form for material that may break its terms of use or the law. It says reported domains may be reviewed by its Model Quality team, which may apply filters or other mitigations so that ChatGPT relies less on unreliable sources.

None of the reporting routes in the table comes with a timescale. Of all the routes in this guide, only Google's knowledge panel feedback has a stated review time.

What not to do

• Do not manufacture mentions. Google's guide to its generative AI features says seeking inauthentic mentions is not as helpful as it might seem, and its spam policies cover attempts to manipulate generative AI responses in Search.
• Do not overstate your own facts. A correction that exaggerates in the other direction is simply a new error for the next answer to repeat.
• Do not promise a date. Treat each route as one input and measure the result, engine by engine.

What to do first

• Save the wrong answer, with the engine, the date, the wording and every cited source.
• Correct the fact on the page that should own it, as plain text, with a date.
• Email the owner of each cited page that repeats the error.
• Report the answer through each engine's feedback control.
• Ask the same question on each engine again after the fixes, and keep asking over the following weeks.