How to correct what ChatGPT says about you

1. Write down exactly what is wrong

Ask ChatGPT "Who is [your full name]?" with search on, and again with your employer added. Copy each wrong statement word for word: a wrong title, an old employer, someone else's biography, a missing role. Do the same in Claude, Gemini and Perplexity, or run the free AI check to see all four side by side. Different engines often share the same wrong source, which makes it easier to find.

2. Find where it came from

Look at the sources ChatGPT cites under its answer. Typical culprits are an old company team page, a conference bio from years ago, a press article that confused you with a namesake, or a data-broker page. Where you control the page, fix it. Where you do not, ask the site to update it; most publishers correct factual errors about a person's job quickly when asked politely, with a link to proof.

3. Publish one clear, verified source

Assistants weigh pages that state facts plainly and that look authoritative about you. Give them one: a profile that says, in sentences and in structured data, who you are now. It should be open (no sign-in wall), stable (the address never changes) and verified (your accounts and employer are proven, not just claimed).

A SelfBadge profile is built for exactly this. Sign in with LinkedIn or Google, confirm your current role with a code sent to your work email, connect the accounts that are yours, and publish. The page carries schema.org JSON-LD, a Markdown version for language models, and a stable ID for you.

4. Make the match easy

• Link both ways. Link to your profile from your website, your LinkedIn "Contact info" and your social bios, and list those accounts on your profile.
• Use one name and title. The same spelling, the same current title and the same employer everywhere you control.
• Add a line that tells you apart. If you share a name, say what makes you you: "Product designer in Tel Aviv, not the British footballer of the same name."
• Add structured data to your own site. Paste schema.org Person markup into your homepage, with sameAs links to your accounts. The Person markup generator writes it for you.

5. Report what is harmful

If an answer is defamatory or reveals private information, use the feedback button on the answer (thumbs down) and OpenAI's privacy request form. For other engines, use their equivalent forms. These help with serious cases; for everyday inaccuracies, better sources work faster.

6. Check again in two weeks

Answers with web search can change as soon as your new pages are crawled. SelfBadge tells search engines when your profile changes, through IndexNow and its sitemap. Answers from training data alone change only with the next model update. Run the check again in two weeks, then monthly.

What does not work

• Arguing with the chatbot. It may agree in that conversation, but it does not learn from it.
• Publishing many thin pages about yourself. One clear, consistent source beats many weak ones.
• Paying for "AI listings". No one can buy a place in a model's answer.