How to fix wrong information about you in ChatGPT and other AI assistants

Why AI assistants get executives wrong

AI assistants build an answer about a person from whatever they can retrieve or remember about that name. Errors usually come from four places: name collisions (other people with the same name get merged into one profile), stale sources (an old company bio or press release still outranks your current role), thin sources (too little about you, so the model fills gaps with plausible guesses), and inconsistent sources (LinkedIn, your company site, and directories each say something slightly different). When I tested my own name in October 2026, one assistant answered "Who is Ben Yeoh?" by describing a Malaysian automotive executive first, and another answer mistook me for a Malaysian building architect. The full results are in the AI executive visibility study (https://benyeoh.com/ai-executive-visibility-study).

Step 1: Record exactly what each assistant says

Before fixing anything, write down a fixed set of questions and run them in ChatGPT, Gemini, Perplexity, and Google (AI Overviews appear on many name searches). Use your full name, your name plus your company, your name plus your city, and the questions a counterpart would actually ask: who you are, what you do, what you have achieved, and what you are known for. Save the date, the exact answer, and every source the assistant cites. Use a logged-out or fresh browser session where you can, because a logged-in account's history can shape the answer you see. This record is your baseline; without it you cannot tell whether a fix worked.

Step 2: Trace each error to its source

For every wrong statement, find where it came from. If the assistant shows citations, open them: the error is often on that page. If it shows no source, the claim probably comes from the model's training data or a page it summarised without citing, which changes more slowly. Group the errors by cause: wrong person, outdated role, missing information, or a third-party page that is simply wrong. Each cause has a different fix, and treating them all as "AI is wrong" wastes effort.

Step 3: Publish one canonical profile you control

Give assistants one authoritative page to prefer: an about page on a domain you own, with your full name, current title, company, city, career history, and verifiable credentials written in plain sentences. Add Person structured data (JSON-LD) with a stable @id, a headshot, your job title, and sameAs links to your verified profiles. Include the details that separate you from people with the same name—city, field, company, notable work—stated as facts about you rather than as comparisons with them. The AI-ready executive profile guide (https://benyeoh.com/ai-ready-executive-profile) covers the markup in detail.

Step 4: Make the sources assistants already trust agree

Assistants lean on a few high-trust sources, so align those with your canonical page word for word: LinkedIn headline and about section, your company's leadership page, professional directories, speaker bios, and any association or regulator listings. In my own test, ChatGPT used LinkedIn to identify me in four of eight questions about me, and for my HRD Corp accreditation it went straight to the HRD Corp trainer directory—then said it could not verify the accreditation, because it could not match me there. If the trusted sources contradict each other, the assistant has to guess; if they agree, it rarely does.

Step 5: Correct or outweigh wrong third-party pages

Ask publishers to update outdated bios, press releases, and event pages—most will for a factual correction. Where a page has already been updated but Google still shows the old version, Google's outdated-content removal request can refresh it. Where you cannot change a page, outweigh it with newer, accurate sources: an interview, a speaker page, or a guide you wrote. For personal data that is wrong, OpenAI and Google provide privacy request forms; expect them to remove information rather than rewrite it. Use the in-app feedback on each wrong answer as well, but treat it as a signal, not a fix. Defamatory content is a legal matter; speak to a lawyer rather than an SEO provider.

Step 6: Re-test on a schedule

Run the same questions every four to six weeks and compare them with your baseline. Answers vary from one run to the next even when nothing has changed: in my own test, the question that cited my site in one run was not the one that cited it ten days later. Look for trends across several runs rather than reacting to a single answer, and re-test after any major change such as a new role, an acquisition, or significant press coverage.

What not to do

Do not create several thin profile sites or near-duplicate city pages; search engines treat them as doorway pages and they add noise rather than clarity. Do not hide text for AI crawlers or try to instruct assistants through invisible prompts on your pages—it is unreliable and can be treated as manipulation. Do not post fake reviews or anonymous articles about yourself. And do not pay anyone who promises to "edit ChatGPT"; no one outside the AI companies can do that.

Three ways to respond to a wrong AI answer

A realistic timeline

How the work usually unfolds for an executive with an existing footprint. Timings depend on how many sources need correcting and how quickly they are recrawled.

Questions