The search answer that said to add glue to pizza

In May 2024, Google started showing AI Overviews, AI-written answers at the top of search results, to everyone in the United States.

Within days, users shared strange answers. One suggested using non-toxic glue to make cheese stick to pizza. Another said geologists recommend eating one rock per day. These appeared to come from Reddit comments and an article from the satirical site The Onion.

On 30 May, Google's head of Search said some odd and wrong overviews had appeared, mostly for unusual questions, and that many screenshots shared online were fake. Google said it had made more than a dozen fixes, including limiting satire and forum posts in answers.

AI Overviews find web pages and then summarise them. The model does not know which pages are jokes. If the only pages about a strange question are satire, it can summarise satire as fact. Google called this a "data void".

The deeper risk is trust. An answer at the top of Google, written in a calm voice, feels checked. It may only be a summary of whatever the system found.

The idea behind this caseLesson: RAG: give AI your documents →

1. 1For health, safety, money or legal questions, open the sources behind the AI answer and judge them yourself.
2. 2Be extra careful with unusual questions: there may be little good information, so the answer may be weak.
3. 3If you build a tool that summarises documents, control which sources it can use.
4. 4Show sources clearly in your own AI tools and teach people to click them.
5. 5Report bad AI answers when you see them, so they get fixed.