Google's "AI Overview" feature is particularly bad.
o3 gives the following response:
> Short answer: it isn’t. Rum isn’t “healthy” any more than other booze.
It goes on to give a bulleted list on why it's not healthy. Full output: https://chatgpt.com/share/6894871f-be9c-800e-8cc9-9e5ec2d5d6...
A sibling comment speculated that Google's response is expected because it's just a summary of bad search results. To test this, I ran a fresh query against o3 with the Web Search option enabled. The results:
> Rum is a tasty way to get ethanol into your body, but it is not a health food. Any modest upsides you may have read about (a bump in “good” HDL cholesterol, slight blood-thinning, a bit of antioxidant pickup from barrel aging) are the same things you’d get from any spirit, and recent large studies show those benefits are either tiny or disappear once you control for lifestyle. Meanwhile the well-documented downsides—cancer risk, liver disease, high blood pressure, weight gain, addiction, injuries—kick in from the first drink.
It goes on to give an extensive debunking of all the nonsense health benefits in the Google summary. It looks like it saw the same search results but was able to course correct and view those results with a more skeptical eye. Full output: https://chatgpt.com/share/6894885d-51dc-800e-ab8c-55af0a5bb2...
This post is about Google, so it's fair to focus on their search product, but I'd be careful to generalize to other LLMs. There are huge differences in quality between different models.