Fake AI Overviews screenshots keep circulating alongside real ones

Screenshots of Google’s AI Overviews answering as a chicken, a fork or a newborn are circulating again, and there is no reporting on any of them.

That absence is the story. The persona format has not been covered by any outlet, so every example in the feed is uncorroborated, and the one thing the record does establish about viral AI Overviews screenshots is that a substantial share of them were never generated at all.

What was documented, and when

The evidence for fabrication comes from the feature’s original rollout in May 2024, not from the current format. Google told Snopes then that many circulating examples of AI Overviews mistakes “were doctored or that we couldn’t reproduce.”

One of the most widely shared of those was posted by Lil Nas X, which NBC News described as “a seemingly fake AI Overview about depression.” Genuine failures existed in the same period, including the advice about using glue to get cheese to stick to pizza, which is why the fakes were credible.

Why that precedent applies to the current format

The 2024 episode established the pattern rather than the specific claims: real errors from the system give fabricated ones cover, and fabricated ones inflate the apparent rate of real failure.

The persona prompts now circulating sit in exactly that gap. They are trivially easy to fake in an image editor, trivially easy to reproduce in a browser, and nobody has done the second thing in public.

What would settle it

Any individual screenshot can be checked by running the prompt, which takes about ten seconds and produces either a matching answer or a discrepancy.

Until someone does that at scale, the 2026 persona trend has the same evidentiary status as the 2024 wave before Google addressed it: a mixture of real output and invention, in unknown proportion, presented as a single phenomenon.