laptop screen showing blue background with blurred keyboard in foreground

When AI Gets Your Name Wrong, Who Puts It Right?

A careful client will look your firm up before the first meeting, and a chatbot may now give the answer. False chatbot answers that confuse one business with another are hard to catch, because nobody tells you what the client read. The answer may describe someone else’s court case or insolvency as yours. You were not there to correct it, and the only sign may be a promising contact who goes silent.

close-up of smartphone screen showing whatsapp messaging app interface

In England and Wales, people tend to treat this either as a passing glitch or as a libel claim waiting to be issued. Both views skip the questions a court would ask first: who saw the answer, what it did to them and where the error came from.

Most people now meet AI answers by default

Firms learned years ago to check the first page of search results for their own name. That habit now covers less ground, because an AI summary often sits above the links. In Ofcom’s 2026 survey of adults’ media use, 75% of online adults said they read AI-generated search summaries at least some of the time. Some 42% read them often or always. As a result, false chatbot answers can reach people who never chose to open a chatbot.

false chatbot answers

The same picture appears elsewhere. Across the UK, 64% of people in an October 2025 survey had seen an AI-generated answer to one of their searches in the previous week. Search results still matter, and plenty of adults never use AI tools at all, so AI checking belongs alongside the old routine. Even so, a business that reads only the links is studying the part of the page its clients may scroll straight past.

Chatbot errors reach readers who trust them

A wrong answer would matter less if readers met it with suspicion. Many do not. Which? put six AI tools through 40 common questions in September 2025, and every tool made repeated factual errors across the 228 answers its experts reviewed. Around half of AI users told Which? they trust the information to a great or reasonable extent, and the share rises to two thirds among frequent users.

Those figures make a mix-up easy to believe. If a tool slips on everyday questions, a question about a small business with a common name gives it more room to go wrong. It might merge two firms with similar names, or give one director the history of another with the same name. The client reading that answer has no reason to doubt it. A credit controller weighing up a new customer may simply decline the account and never say why.

Proving harm from false chatbot answers is hard

The law may already reach this problem. On 7 July 2026, the UK Jurisdiction Taskforce published a statement on liability for AI harms concluding that existing doctrines, including defamation and negligent misstatement, already provide a framework for who is responsible. For a business on the wrong end of a mix-up, the ordinary tests therefore apply. Under section 1 of the Defamation Act 2013, a claimant must prove serious harm, which means facts showing the material impact of the statement.

tablet and printed papers spread on a wooden office desk

An American case shows how a claim can founder on thin evidence of harm. In May 2025, a Georgia court ended Walters v OpenAI (Case No. 23-A-04860-2) without a trial, granting summary judgment to OpenAI. It held that no reasonable reader would take the output as fact, and that the claimant had shown neither fault nor recoverable damages. The false statement had gone to one person, who recognised it as wrong. American law differs, and no English court has yet decided how such a claim would fare.

Tracing the source is where correction starts

The regulator has seen the risk coming. In April 2024, the Information Commissioner’s Office warned that wrongly relying on generative AI for facts about people can lead to “misinformation, reputational damage and other harms”. An individual confused with someone else can use UK GDPR and the Data Protection Act 2018 to ask for false personal data to be rectified or removed. A business needs a second lever, and the best one is the material the tool drew on.

false chatbot answers

Chatbot errors may echo something published elsewhere, such as an old article about a similarly named firm or a directory entry that merges two companies. Reputational risk work now includes capturing the answer, repeating the question across tools and dates, and finding the pages behind it. The business can then ask each site owner to correct the error where it started. Where several publications say similar things, a claimant may find it harder to show which one caused the harm.

Some say firms should go straight to court

A litigator might call all this too cautious. Every day spent gathering evidence is another day a falsehood stays live, and the law needs test cases to move. No UK court has yet decided whether a chatbot operator counts as a publisher. The UK Human Rights Blog reported in November 2025 that a Belfast libel lawyer is preparing a group action, arguing that generative output is new material outside the section 5 defence for intermediaries. On that view, businesses hit by false chatbot answers should join the fight.

That view deserves respect, and the law may well move its way. Even so, the action had not been issued when it was reported, and any claimant would still have to prove serious harm. Walters, a Georgia decision on American grounds, shows a claim can fail when the single recipient recognised the answer as wrong and no damage was shown. Every claim begins with proof of what was said, to whom and from which source, and a correction request needs the same file.

Checking for false chatbot answers becomes routine work

Go back to the client who read the wrong answer before the first meeting. A firm that checks routinely may have found that answer weeks earlier and traced it to a page about another business. Who the law will hold responsible remains open. Because AI has no legal personality, the UK Jurisdiction Taskforce statement says liability attaches to legal persons such as developers, deployers and users. Clear warnings that words are AI-generated may also affect what the words mean and whether the serious harm threshold is met. A firm that wants to be ready should start a dated record of what each tool says about it.

The routine is simple. At set intervals, ask the main AI tools the questions a client would ask about the business and its directors. Record the date, the tool, the prompt and the full answer. Where inaccurate AI answers appear, trace the pages behind them, ask site owners to correct them and, for an individual, consider a data protection request. If an answer is damaging and the source stays uncorrected, take legal advice with that file in hand.

If you have found false chatbot answers about your business or its directors, or want a routine check of what AI tools say about you, Tremark Associates can help. Call our team on 0113 263 6466 or use the form below to discuss how we can trace where a wrong answer came from.

    Are you a private individual or a business?

    Frequently Asked Questions

    What can a business do about false chatbot answers?

    Note the date, the tool, the exact question and the whole answer, then ask the same question again later and on other tools. Look for the web pages carrying the wrong information and ask their owners to correct them. If the answer is damaging and stays uncorrected, a solicitor can advise on the options using that file.

    Can an AI tool’s wrong statement about a business be libel in England?

    Possibly, though no UK court has yet decided the point. Because AI has no legal personality, any claim would have to be brought against a person or company. That could be the developer, the business deploying the tool or a user who repeated the words. The claimant would still need to prove serious harm, which is difficult where few people saw the answer.

    Can an individual ask for false AI-generated personal data to be corrected under data protection law?

    Yes. The ICO has consulted on how the accuracy principle applies to the outputs of generative AI models. A request backed by a clear, dated record of the wrong answer gives the recipient something specific to act on.

    Categories

    • Guidance

    Popular Blogs