How to Fix Wrong Facts About Your Brand in ChatGPT, Gemini and Perplexity

To fix a wrong fact about your brand in an AI engine, capture the exact answer, find the source it relied on, correct that source (your site first, then third-party pages and profiles), report the error in the product, and re-test after recrawl. Engines that search the web repeat what they find, so the fix is upstream.

Examples are illustrative, not MentionX measurements.

Why do engines get brand facts wrong?

• Outdated sources. An old pricing page or a 2023 review still ranks.
• Conflicting sources. Your site and a directory disagree.
• Training data. Without search, a model answers from what it learned, which can be out of date. OpenAI notes ChatGPT may give an initial answer before considering search results (OpenAI).
• Misreading. The engine merges you with a similarly named company.

Step 1: How do you capture the error?

1. Record the prompt, engine, date, and whether search was used.
2. Copy the full answer and any cited sources. ChatGPT shows inline citations and a Sources panel (OpenAI); Gemini shows a Sources button when sources are available (Google); Copilot shows the exact Bing query and its sources (Microsoft).
3. Re-run the prompt 3–5 times to see if the error is consistent.

Step 2: How do you find the source?

• Cited answers: open each citation and find the wrong claim.
• Uncited answers: search the claim in Google and Bing; the top results are likely sources.
• Copilot: use the shown query to see what Bing returns (Microsoft).
• No source found: the error probably comes from training data; focus on making correct facts prominent and consistent.

Step 3: How do you fix the source?

Also make the correct fact easy to extract: state it plainly in text, near the top of the relevant page.

Step 4: How do you report it in the engine?

• ChatGPT: use thumbs-down feedback on the response; for product results, report incorrect information from the product menu or the content reporting form (OpenAI).
• Gemini and AI Overviews: use the in-product feedback controls.
• Perplexity: use in-product feedback.
• Copilot: use in-product feedback.

Feedback helps, but there's no guarantee of a specific change. The source fix is what lasts.

Step 5: How do you re-test?

1. After recrawl (check the page's last crawl date), re-run the same prompts.
2. Log whether the claim changed and what is now cited.
3. If the error persists with search on, look for another source.
4. If it persists only with search off, it's likely training data. Keep correct facts consistent and wait for model updates.

What are engine-specific notes?

What should you avoid?

• Publishing many pages repeating the fact (looks like spam (Google))
• Arguing with the engine in chat and assuming it "learned"
• Fixing only your site while an outdated third-party page still ranks

Checklist

• [ ] Error captured with prompt, date, engine and sources
• [ ] Source identified
• [ ] Own pages updated and recrawl requested
• [ ] Third-party corrections requested
• [ ] Feedback submitted
• [ ] Re-test logged

Related

• Why AI engines ignore your brand: 7 diagnoses

Sources

• OpenAI, Searching the web with ChatGPT; Shopping with ChatGPT search; OpenAI crawlers
• Google, Gemini sources; AI features and your website; Block indexing; Merchant Center; Spam policies
• Microsoft, Copilot web search; Manage public web access; Bing, AI Performance
• Perplexity, Perplexity crawlers; Wikidata, Introduction

All accessed 30 Oct 2026.