Google's AI falsely brands blameless hospital worker a convicted drunk driver

What happened

Tracey Melsom-Page has worked as an anaesthetic technician in West Auckland, New Zealand, for 39 years. She has held three senior roles in public hospitals and works closely with surgeons and anaesthetists. By her own account, she has never been in trouble with the law, apart from the odd speeding ticket. But anyone who typed her name into Google saw something different. According to Google's "AI Overview", she had been convicted of drink-driving and disciplined by the Health Practitioners Disciplinary Tribunal for lying about her convictions. The tribunal is New Zealand's disciplinary body for health professionals. The AI Overview is the AI-generated summary Google shows above the regular search results.

None of it is true. Melsom-Page has no criminal record and is still allowed to practise. The case came to light through a New Zealand Herald report published by Newstalk ZB. The Melsom-Pages had turned to the paper in desperation.

How the AI went off track

About three months ago, a colleague searched for her name online. She came across a link to a 2023 Herald article about an anaesthetic technician who faced disciplinary proceedings over drink-driving and giving false information. The article, however, didn't name the person involved and never mentioned Melsom-Page.

How the AI linked her name to the case is unclear, even to her. What stands out most is what the Herald documented. When the paper pushed back, the AI initially doubled down. It claimed that tribunal records named Melsom-Page and that it had cross-checked this against official databases. Only after further questioning did the system admit a "logical error": the decision it had cited showed no connection to her at all.

Experts call errors like this "hallucinations": a language model produces statements that sound plausible but are false. It has no real knowledge; it strings words together based on what's statistically likely. In this case, the job title and region matched. The AI then apparently made up a link to a real person on its own, and on top of that claimed it had checked sources that confirmed it.

"It's ruining my life"

For her, the consequences are serious.

"It's ruining my life. And it feels like there's nothing I can do about it."

What worries her most is what colleagues, surgeons and potential future employers might believe. "People who know me know I didn't do it, but it's quite hurtful and hard," she told the Herald. Thirty-nine years in the profession and everything she has built up, she said, and it's "just devastating".

Her husband thinks of their grandchildren: if they google their grandmother one day, that's exactly what they'll find. The couple describe themselves as "cyber hermits". Melsom-Page avoids social media and deliberately keeps her private life offline. That doesn't help, though, against an AI that draws its own connections from other people's content. "AI is now taking over my life for the future. What's that going to look like in five to ten years?" she asks.

Three hours of forms, no result

The couple got support from Netsafe, New Zealand's online safety agency, which advised them to contact Google directly. Last weekend, they spent more than three hours on online forms and links, some of which didn't work. They got nowhere. Her husband called the whole thing "pathetic". Only then did they go to the press.

Asked by the Herald, Google said it was investigating the false association and would fix the problem. A spokesperson said the "vast majority" of AI Overviews provide accurate information, but the technology can sometimes get things wrong:

"When our features misinterpret web content or miss important context, we take action and use those examples to improve our systems."

The Herald also asked whether Google would formally apologize to Melsom-Page or compensate her. The company hadn't responded by the time of publication.

Not an isolated case in New Zealand

The case echoes another one from New Zealand. Businessman Wayne Wright jnr, who appears on New Zealand's rich lists, was confused by Grok, the AI chatbot from Elon Musk's company xAI, with a man convicted of possessing child sexual abuse material. The actual offender had permanent name suppression: in New Zealand, courts can order that a defendant's name must not be published. Grok linked Wright to the case anyway. Wright contacted Musk directly to get it corrected. He also called on the real offender to come forward so that no more innocent people would be accused.

Both cases have something in common. The original reports deliberately left the person unnamed, either for editorial reasons or because a court ordered it. And it's precisely that gap the AI fills with a real but wrong name.

What you can do if it happens to you

Sean Lyons, Chief Online Safety Officer at Netsafe, sees this as a new form of harm. AI can analyze huge amounts of data and in the process draw connections between people and events that seem logical but are false. Systems that may have been built without adequate safeguards and cause such harm "autonomously" should concern everyone, he said.

Netsafe has so far received only a few complaints specifically about AI misinformation. But Lyons suspects the problem is more widespread than many realize. He's also aware how hard it can be to get tech companies to correct things: if someone hits a brick wall, that's "not good enough".

Lyons points to the Harmful Digital Communications Act, a New Zealand law against harmful digital content. It applies regardless of whether a human or an AI created the content. What matters is that someone suffers serious emotional distress because of digital content. The District Court has previously ordered search engines to remove links connecting people to harmful material. People affected should contact Netsafe even if they don't know how the false information came about.

The bigger picture

This is more than an embarrassing glitch. AI Overviews sit at the very top of the world's most-used search engine. Many people read only the summary and never click through to the sources. That turns a hallucination into a public statement of fact about a real person. The fact that the AI, when challenged, even invented a supposed check of official databases makes it worse, because users expect evidence from a search engine, not made-up confirmations.

The case also exposes a structural problem: there are hardly any ways to complain. If you're wrongly accused, you end up stuck in forms. Google only responded once a newspaper asked questions. It's still unclear whether the false claim has since disappeared, how often it was shown, and how it came about in the first place. Also unresolved is who is liable when an AI makes damaging claims on its own. New Zealand at least has a legal lever in the Harmful Digital Communications Act, and similar cases are keeping courts busy in other countries too. For Google, pointing out that the "vast majority" of overviews are accurate is a weak argument when a single wrong answer can put someone's livelihood at risk.