Google's AI Overviews Are Answering Millions of Questions Wrong, and Bluesky Has Stopped Pretending It's a Small Problem

A post on Bluesky put it simply: "Google's AI Overviews are peddling misinformation on a scale that may be unprecedented in human history."[¹] The post got 45 likes — modest by viral standards, but it was one of dozens making the same claim in the same 48-hour window, all pointing to the same analysis, all using variations of the same phrase: unprecedented. When a community starts reaching for superlatives in unison, it's worth asking what broke the dam.

The proximate trigger was a Futurism analysis finding that Google's AI Overview feature was generating wrong answers at a rate so high that the error volume, multiplied across the billions of queries Google handles daily, dwarfs anything previous misinformation researchers had to contend with. One post that drew significant engagement cited a supporting statistic that has become the conversation's sharpest edge: only 8% of users actually verify what an AI tells them.[²] That number does more damage than any volume estimate, because it reframes the problem from "AI makes mistakes" to "AI makes mistakes that almost no one catches." The AI misinformation conversation has been building toward this framing for months — the earlier debate over whether AI systems could generate fictional diseases and present them as real now looks like a preview of a much larger argument.

What's notable about the current moment is how little defense the Google Overviews product is getting, even from people who are usually skeptical of AI panic. One Bluesky commenter, who had been mocking the "AI crowd" for treating technical complaints as misinformation, found themselves at the center of a pile-on — their joke about a site's outage was called misinformation by other users, which they experienced as absurd overreach.[³] The exchange captures something real: the word "misinformation" has become so freighted in this community that it now functions as both a serious accusation and a social weapon, and people are confused about which one they're receiving. That confusion is doing real damage to what could otherwise be a productive conversation about verification and trust.

The news coverage running parallel to the Bluesky conversation is almost entirely about fraud — AI-powered identity theft, deepfake schemes targeting financial institutions, North Korean IT workers using synthetic faces to pass security checks. This is misinformation as operational infrastructure, not as accidental error, and it sits in an entirely different register from the Google Overviews debate. The two conversations rarely touch, which is a problem: the companies building AI search features and the criminal organizations exploiting generative AI for fraud are working from the same underlying capabilities, but they're being discussed in separate editorial silos. Fintech trade press runs its AI fraud warnings; Bluesky users share their AI Overviews horror stories; and nobody is connecting the systems.

The thread running through all of it is trust calibration — or rather, its failure. The 8% verification figure isn't an anomaly. It reflects something researchers have observed repeatedly: people extend to AI systems a default credibility they wouldn't give a random website. That credibility was built, in part, by Google itself, which spent two decades training users to treat its search results as authoritative. Now Google has inserted a layer that can be confidently wrong, and the epistemic habits it cultivated are working against the very users it's supposed to serve. The Bluesky community has reached a verdict on this — and their verdict is that Google created the problem it is failing to fix. The more interesting question, which the current conversation hasn't quite reached, is what it would actually take to rebuild verification habits at scale. On current evidence, not much.