Google’s AI Overviews were meant to make search results easier to digest. Instead, a handful of strange answers became a public test of how much trust users should place in AI-generated summaries at the top of search.
After bizarre examples spread widely on social media, Google’s head of search, Liz Reid, acknowledged that the rollout exposed areas that needed work. In a post published late Thursday, she wrote, “We wanted to explain what happened and the steps we’ve taken.”
What went wrong with AI Overviews
The most visible failures were not subtle. One AI Overview appeared to endorse eating rocks because doing so “can be good for you.” Another suggested using nontoxic glue to thicken pizza sauce.
Those examples mattered because AI Overviews are not just another link in search results. They are presented as a synthesized answer, which can make an error feel more authoritative than a strange forum post or a satirical article buried lower on a page.
Reid’s explanation pointed to different causes for the two viral mistakes. In the rock-eating case, the problem began with a topic that few people seriously discuss online. Because there were not many useful sources for a search engine to draw from, the system found an article from The Onion, a satirical website, after it had been reposted by a software company. The AI tool then treated the information as if it were factual.
The pizza example showed a different weakness. Reid wrote, “We saw AI Overviews that featured sarcastic or troll-y content from discussion forums.” She added that forums can often provide authentic, first-hand information, but in some cases they can also lead to advice such as using glue to get cheese to stick to pizza.
Why viral screenshots made the story harder to judge
Google also argued that the public conversation around AI Overviews was shaped by screenshots that did not all reflect real search results. According to the source article, some widely shared images of failed AI Overviews were fake, and WIRED’s own testing supported that conclusion in at least one case.
One example involved a post on X that appeared to show an AI Overview responding to the question “Can a cockroach live in your penis?” with an enthusiastic confirmation that this is normal. The post was viewed over 5 million times. But the screenshot format did not match how AI Overviews are actually shown to users, and WIRED was not able to recreate a similar result.
The confusion extended beyond social media users. The New York Times issued a correction to reporting about the feature, clarifying that AI Overviews never told users experiencing depression to jump off the Golden Gate Bridge. That claim, according to the source article, came from a dark meme on social media.
Reid also pushed back on other claims. “Others have implied that we returned dangerous results for topics like leaving dogs in cars, smoking while pregnant, and depression,” she wrote Thursday. “Those AI Overviews never appeared.”
That distinction is important, but it does not erase the real errors. Fake screenshots can exaggerate the problem, yet the authentic examples still show how AI search can mishandle satire, sarcasm, and weak source material.
Google’s defense of the feature
Reid argued that judging AI Overviews by viral screenshots alone would be unfair. She said Google conducted extensive testing before launch and claimed that the company’s data shows people value AI Overviews. One sign she pointed to was that people are more likely to stay on a page discovered through the feature.
She also framed the wave of attention as a stress test driven by large-scale public use. “There’s nothing quite like having millions of people using the feature with many novel searches,” she wrote. “We’ve also seen nonsensical new searches, seemingly aimed at producing erroneous results.”
That explanation gives two views of the same launch. On one hand, a product used by millions will inevitably face unusual queries and attempts to make it fail. On the other hand, a search feature that summarizes information for users has to be especially careful when the available source material is thin, sarcastic, or misleading.
What Google says it changed
Reid said Google made “more than a dozen technical improvements” to AI Overviews after the viral failures. The post described only four of them.
- Better detection of “nonsensical queries” that should not receive an AI Overview.
- Less reliance on user-generated content from sites like Reddit.
- Showing AI Overviews less often in situations where users have not found them helpful.
- Stronger guardrails that disable AI summaries on important topics such as health.
These changes target the specific weaknesses highlighted by the rollout. If a query is designed to produce a strange answer, Google says the system should be better at avoiding an AI-generated response. If the available material comes from forums or other user-generated sources, the feature should rely on that material less heavily. And for sensitive topics, especially health, the company says guardrails have been strengthened.
The bigger search problem
The episode shows how AI-generated search answers can fail in ways that ordinary search results do not. Traditional search can surface a bad source, but users still see the source as a separate result. AI Overviews can blend information into a single response, which means the system’s judgment about source quality becomes part of the answer itself.
The rock example is especially revealing. The issue was not simply that satire existed online. It was that the AI Overview treated a satirical claim, reposted elsewhere, as usable information for a real answer. The pizza example shows the same problem from another angle: discussion forums can be useful, but jokes and sarcasm can become dangerous when stripped of context.
Google is not saying it will significantly roll back AI Overviews. According to the source article, Reid’s blog post did not mention a broad retreat from AI summaries. Instead, Google says it will keep monitoring user feedback and adjust the feature as needed.
For users, the practical lesson is straightforward. AI Overviews may be useful, but they still require scrutiny, especially when an answer sounds strange, involves health or safety, or appears to come from thin online evidence. The feature is changing, but the need to read carefully has not gone away.