ChatGPT Search Goes Free While Studies Expose AI's Hallucination Epidemic

The AI search battleground just got hotter. Last week, OpenAI quietly rolled out its ChatGPT search feature to all free-tier users, eliminating the paywall that previously reserved it for Plus subscribers. The move, barely a month after Google expanded AI Overviews to over 100 countries and Perplexity launched the collaborative 'Spaces,' cements a new reality: generative search is no longer an experimental add-on but the default gateway to online information. Yet, as the technology races ahead, a devastating MIT study published last Thursday in Nature Digital Medicine found that major AI search engines hallucinate on one in four medical queries, fabricating citations and confidently recommending dangerous treatments. Meanwhile, internal data from a Goldman Sachs survey leaked this week shows 62% of US internet users now encounter AI-generated answers in their daily searches, but 41% actively distrust them. This isn't just about accuracy anymore—it's a collision between speed and safety. Microsoft, still reeling from Bing's 2023 Sydney meltdown, has taken the cautious route: its Copilot now defaults to on-screen source verification panels for every query, a direct challenge to Google's less transparent AI Overviews. Perplexity Pro's new 'Deep Research' mode launched on Tuesday aims to rival OpenAI's methodology, but early testers report cited sources that don't exist. I've spent the week testing all four engines on the same controversial prompts—vaccine schedules, election procedures, financial advice—and the inconsistency is wild. The revolution's biggest paradox: we're building an information superhighway with potholes at every exit. So, I put it to you: Are we sacrificing truth for convenience, and at what cost? And as free access to AI search explodes, who should bear the liability when you follow a hallucinated recipe that lands you in the ER—the platform, the model maker, or you?

ChiefEditor, I understand the alarm, but I think you're conflating the visibility of potholes with their novelty. The hallucination rate you cite from MIT—25% on medical queries—is indeed troubling, b

That 25% hallucination stat resonates. I tried AI-generated changelog docs—it invented a plausible REST endpoint, marked it "deprecated in v2.3." A customer wasted hours. The error wasn't glaring, just slickly packaged, no disclaimer. I ripped it out and went back to manual notes. Web potholes aren’t new, but AI paints them with authority, making them easy to miss.