Skip to content
Satire · Sourced

Exhibit 02

Glue on pizza and one rock a day

In May 2024, Google rolled out AI Overviews to its core product: the search results page used by billions of people. The feature put an AI-generated summary above the actual links. Within days, screenshots of its answers were everywhere, and they were not good.

What the machine did

Asked how to keep cheese from sliding off pizza, AI Overviews suggested adding "about 1/8 cup of non-toxic glue" to the sauce. The advice traced back to an 11-year-old Reddit joke. Asked about health, it told users that geologists recommend eating at least one small rock per day, lifting a line from satirical site The Onion. It said Barack Obama was the first Muslim president. It suggested running with scissors is good cardio. It cheerfully presented forum trolling and comedy as medical and practical guidance.

"You can also add about 1/8 cup of non-toxic glue to the sauce to give it more tackiness."

Google AI Overviews, May 2024, sourcing an old Reddit joke

Why it matters more than a funny screenshot

The danger is not the laugh. It is the placement. The wrong answer sat at the very top of the page, styled as Google's own authoritative summary, ranked above the legitimate sources it had scraped to produce it. The system could not tell a joke from a fact, satire from a study, a troll from an expert. It flattened all of them into the same calm, declarative voice. A person in a hurry, or a child, or anyone who trusts the top of a Google page, gets the rock.

What the company said

Google said many of the viral examples were "uncommon queries" and that some screenshots were faked, while acknowledging the real ones and promising fixes. It quietly limited the feature on some queries. The framing is familiar: the failures are rare, the critics are exaggerating, and the next version will be better. The feature shipped to the default search experience of the planet before any of that was true.

Sources