AI tools need better receipts

A BBC investigation found that AI chatbots can be manipulated into giving false or biased answers, sometimes from just one online post.

The issue happens when tools like ChatGPT, Gemini, Claude, and Google’s AI Overviews search the web for answers.

If they pull from a weak or misleading source, they can repeat that information as if it is reliable.

To test this, BBC journalist Thomas Germain published a fake article claiming he was a world-champion hot-dog eater.

Within a day, major AI tools were repeating the claim. The example was funny, but the problem is serious.

The same tactic has reportedly been used around health supplements, financial advice, and retirement information, where bad answers could cause real harm.

Experts say AI search makes this risk bigger because it often gives users one direct answer instead of several links to compare.

That makes it easier to trust the response without checking where it came from.

Google has updated its spam policies to make clear that attempts to manipulate AI answers break its rules.

The company says this is only a clarification, not a major change.

Still, experts believe Google and other AI companies are trying to add more safeguards.

Some tools now appear to be adding more warnings, caveats, or confidence labels to AI answers.

But experts say bad actors may simply move to other tactics, such as using videos, influencers, or multiple websites to push the same claim.

For now, AI answers should be treated as a starting point, especially for health, money, or legal advice.

Are we really back to glue on pizza in 2026? C’mon guys. - MV