The AI Overviews Get a Crowdsourced Upgrade
Google is doubling down on its AI Overviews by injecting them with quotes from Reddit, blogs, and other web forums. The company claims this move will help users find niche advice by surfacing “perspectives from public online discussions.” But this feels less like innovation and more like admitting that its AI has failed to generate trustworthy answers on its own. Two years after its disastrous launch, where the AI famously advised eating rocks and putting glue on pizza, Google is outsourcing its credibility to the very platforms users already bypassed it to reach.
Why This Feels Like a Step Backward
The core problem remains: LLMs hallucinate, and now Google is adding crowdsourced noise on top of that. A 90% accuracy rate sounds impressive until you realize it means hundreds of thousands of Google searches per minute return garbage. Google is effectively blurring the line between an answer and a search results page. If the AI Overview is just a curated list of forum posts with usernames attached, what separates it from a regular search? The company is desperately trying to keep users from typing “Reddit” after every query, but this strategy risks training users to trust anonymous rants and unverified hot takes.
The Trust Fallacy
Google is dressing up this change with more context, showing creator names and community tags. It’s a similar tactic to what ChatGPT and Claude do with citations. But those systems often hallucinate the validity of their sources. Google is now playing the same game, but with the entire web as its hallucination fuel. Double checking your AI’s work shouldn’t be mandatory, yet here we are, stuck in a feedback loop where the search engine is just an aggregator of its own worst content.
Source: Techcrunch
