What Happens When AI Chatbots Talk About You: Personal Reputation Management in the LLM Era

That is a stranger’s judgment of you, delegated to a model, before a second conversation even happens.

Hiring works the same way, just with better documentation. Recruiter behavior has been consistent for a decade: seven in ten companies conduct social media background checks on candidates, and nearly all businesses do some form of online research before making an offer. What changed recently is the tool doing the research. A 2026 industry review found AI adoption among HR professionals climbed from 58% in 2024 to 72% in 2025, with recruiters increasingly asking a chatbot to summarize a candidate’s online presence instead of reading it themselves. The summary a hiring manager reads may never mention that it came from a five-year-old forum post or a name collision with someone else entirely.

The through-line: your name is being asked about by people who will never tell you they asked, and the answer they get shapes a decision before you’re in the room.

Radio host Mark Walters sued OpenAI in 2023 after ChatGPT told a journalist, with total confidence and zero basis, that Walters had been criminally charged with embezzling from a nonprofit. Nothing about it was true. The court ultimately sided with OpenAI, partly because the journalist never republished the false claim and partly because ChatGPT’s own disclaimers about accuracy undercut the case for reliance. But the pattern the case established — a model inventing a criminal accusation and presenting it as fact — did not go away.

A more consequential case is still moving. Conservative activist Robby Starbuck sued Google over Gemini and Bard outputs that allegedly fabricated sexual assault accusations, invented criminal records, and cited court documents that don’t exist. In July 2026, a Delaware Superior Court judge denied Google’s motion to dismiss, letting the case proceed to discovery on more than twenty separate AI-generated outputs about him. That’s a meaningfully further stage than Walters ever reached, and it’s currently the clearest signal courts are willing to give an individual’s claim against a chatbot a real hearing rather than dismissing it on the disclaimer alone.

Neither case is a settled win for the plaintiff. Both point to the same uncomfortable fact: the legal system is still deciding whether being defamed by a hallucination counts the same as being defamed by a person, and until that’s resolved, an individual whose name gets fabricated by a chatbot has real but unreliable recourse. Waiting on a lawsuit to fix what a model says about you is a multi-year bet with thin odds. That is the argument for building an accurate record before the model needs one, not after.

Regulators are further along than courts on at least one piece of this. In April 2024, the European privacy group noyb filed a GDPR complaint against OpenAI after ChatGPT repeatedly generated an incorrect birth date for a public figure and OpenAI told the complainant it had no way to isolate and correct that single fact without affecting other information the model generated about him. Under GDPR Article 17, EU residents have a right to have inaccurate personal data corrected or erased. Whether a large language model can honor that right at the level of one fact about one person remains an open technical and legal question, and it is exactly the kind of gap that leaves an individual’s name exposed regardless of which jurisdiction’s court eventually rules.

A chatbot synthesizing an answer about you doesn’t wait for permission. It pulls from whatever exists: an old court record, a mistaken-identity mention on a forum, a competitor’s negative blog post about someone with a similar name, a data broker profile nobody asked to be created. If nothing accurate and substantial exists to balance that against, the model’s answer is built entirely from whatever it found, and it delivers that answer with the same confident tone it would use for something true.

Silence isn’t neutral here. It’s a vacuum, and vacuums get filled by whoever else is talking about you — or by nothing at all, which is its own problem when a client or employer expects to find something and finds an empty search result instead.

1. Baseline audit: Search your name across Google, Bing, and image results, then ask ChatGPT, Gemini, and Perplexity directly what they know about you. Screenshot everything. You can’t fix an answer you haven’t seen, and most people have never actually looked.
2. Third-party record building: A personal website telling your own story carries less weight with a model than independent coverage does, for the same reason a company’s own press release carries less weight than a journalist’s reporting. Interviews, guest contributions to publications in your field, speaking engagements, and professional profiles on platforms a model treats as credible all function as the counterweight to whatever negative or mistaken content might otherwise dominate the source set.
3. Structured data on anything you control: Person and Organization schema on your own site, a complete and accurate LinkedIn profile, consistent name formatting across every platform. This is like giving a model a fact it can lift cleanly instead of a paragraph it has to interpret, and interpretation is where errors get introduced.
4. Review and people-search cleanup: Data broker listings, outdated employer information, and mismatched biographical details are exactly the kind of noise that produces a wrong birth date, a wrong job title, or a wrong city — the small factual errors that seem harmless until a model repeats one as an anchor fact in a larger answer.
5. Recurring monitoring, not a one-time check: AI-generated answers regenerate every time someone asks, so a problem fixed once can resurface if the underlying source set shifts. A quarterly check of what the major assistants say about your name catches drift before a client or employer does.
6. A documented correction process: When something is genuinely wrong, know in advance how to request a correction — through the platform’s own feedback mechanism, through a GDPR request if you’re in the EU, or through a firm that handles this professionally. NetReputation and similar agencies increasingly treat this exact workflow- auditing what search engines and AI assistants currently say about a person’s name before any content work begins- as the starting point for individual clients the same way they’d start a corporate engagement.

None of this guarantees a model never gets something wrong about you. Nothing does, yet, including a lawsuit. What it does is stack the source set in your favor, so when a model synthesizes an answer, the material it draws from reflects who you are.

That asymmetry — someone else learns something about you, and you never find out what — is the actual argument for treating personal reputation management as ongoing infrastructure instead of a one-time cleanup. The people who get ahead of it aren’t doing anything complicated. They’re checking what the models say, building the record that gives those models something accurate to work from, and treating their own name with the same discipline a smart company applies to its brand.