Google AI Overviews put people at risk of harm with misleading health advice

https://www.theguardian.com/technology/2026/jan/02/google-ai-overviews-risk-harm-misleading-health-information

by sandebert • 7 months ago

42 10 7 months ago

brushfoot• 7 months ago

These AI Overviews are awful. I've been documenting the ones I've gotten over the past few months. Examples:

- 2025-09-19. My query: "is mics an abbreviation for micrograms." AI Overview: "No, MICs is not an abbreviation for micrograms; it is an abbreviation for Minimum Inhibitory Concentration."

- 2025-09-19. My query: "75 mics of medication." AI Overview: "When discussing medication, 'mics' is a common abbreviation for micrograms (mcg)."

- 2025-11-03. My query: "copilot 'replace string in file'." AI Overview: "While Copilot in tools like Visual Studio Code can assist with code generation and refactoring, its primary function is not directly to perform 'replace string in file' operations across an entire project." ("Replace string in file" is the name of an operation that Copilot performs, and I was looking for more info about how it works.)

- 2025-11-22. My query: "u2 'spirits move you'." AI Overview: "The phrase 'spirits move you' is not a direct U2 song title, but it likely refers to their song 'With or Without You,' a famous track from their album The Joshua Tree." (Who said anything about it being a "direct U2 song title"? It's a lyric from "Mysterious Ways.")

It's so frequently wrong and so frequently makes insulting assumptions that it's worse than worthless. And when you click the "Dive deeper in AI mode" button at the bottom, the new response often contradicts the old one. Just garbage.

halfmatthalfcat• 7 months ago

Welcome to Slopworld

duskwuff2• 7 months ago

Not mine, but:

Query: "max amps 22 awg"

AI Overview: "A 22 American Wire Gauge (AWG) copper wire can carry a maximum of 551 amps. Here are the fusing currents for some other AWG wire sizes..."

(A fusing current is the current that will instantly melt and possibly even vaporize a wire. The safe operating limit is two or three orders of magnitude lower.)

londons_explore• 7 months ago

I mean it isn't technically wrong...

Sometimes you need to know how much current is guaranteed to vaporize a bit of wire - for example if you are designing a fusible link.

Gorkys• 7 months ago

I asked

Is a panda or a Clio bigger

Would have thought it was obvious I was talking about cars? Its reply was hilarious:

A Renault Clio is much bigger than a giant panda. While a panda can be 4 to 6 feet long and 2 to 3 feet tall, a Clio is typically 4.053 meters long and 1.440 meters tall, making it significantly larger in all dimensions, according to Auto Express and CarsGuide.

therobots927• 7 months ago

I Googled “can X and Y medications be combined” today actually and to my surprise was provided with an AI overview. It took the cautious route and recommended they not be combined, but I’m surprised that AI overviews for this type of thing would be allowed. In the event that someone were incorrectly advised to combine X and Y and died as a result, would it be possible for a lawyer to subpoena the search history of the individual and hold Google liable for giving incorrect medical advice?

Edit: Expanding the thought process here, if the DOJ wanted to could they force Google to release records of instances when such health-related queries were made, and instances where that was the last thing the person had googled since?

greatgib• 7 months ago

The problem is not the search or some AI generated bullshit.

The problem is that we need to teach again to people that not everything they read from internet or AI is true. Starting at the young age!

kotaKat• 7 months ago

> A Google spokesperson said that many of the health examples shared with them were “incomplete screenshots”, but from what they could assess they linked “to well-known, reputable sources and recommend seeking out expert advice”.

Do Google employees get expert instruction on how to gaslight people, or does that come naturally from their dating lives?

Google AI Overviews put people at risk of harm with misleading health advice

Related Stories