This is a direct reaction to an investigation by The Guardian (yay, media can still influence corporations). Journalists found that for queries like "normal liver levels," the model was producing incorrect information that contradicts medical standards. In simpler terms — complete nonsense.
Google only removed the overviews after the problem became public. Officially, they say the system is dynamic and learns from feedback, but in reality — it's manual firefighting.
The block works not conceptually, but by keywords. If you ask "what is the normal range for liver blood tests," the AI is silent. But slightly change the wording (for example, to "lft reference range") — and generation returns, along with the risk of getting a hallucination instead of facts. I checked. In Russian, Overviews still work with any medical query.
Bottom line: using LLMs as a doctor replacement is still premature. This isn't code refactoring — the cost of an error here is much higher.
Source from The Guardian: https://www.theguardian.com/technology/2026/jan/11/google-ai-overviews-health-guardian-investigation