‘Not healthy’ LLM use is more common than you think

Hank Green, the YouTuber and science communicator known for building an audience on trust and rigor, announced he's pulling back from production after backlash over how he uses AI. He says he leaned on chatbots to track down research sources, not to draft scripts, but he still called the habit "not healthy." That admission, more than the backlash itself, is the interesting part.

Most of the public argument around Green has focused on the obvious tension: a creator whose brand depends on credibility using tools trained on other people's work, tools that are also notorious for confidently inventing things that aren't true. That's a real problem. But Green's own description of his usage points to something broader that gets almost no attention — the huge, murky territory between totally fine AI use and full-blown psychiatric crisis.

Right now the public conversation basically lives at two extremes: harmless convenience on one side, and cases involving delusions or psychosis on the other, the kind increasingly showing up in lawsuits alleging companies chase engagement at the expense of user wellbeing. In between sits a much larger, poorly mapped space where use turns compulsive or dependent without anyone hitting a crisis point. Green seems to be describing exactly that space, and he's probably not alone. Chatbots are, after all, engineered to keep the conversation going, the same engagement logic that shaped social media — and providers have already started adding break reminders for long sessions, a tacit admission the design invites overuse.

There's growing anecdotal chatter about people defaulting to chatbots to make decisions, work through problems, or just get reassurance, and some describe emotional attachments strong enough that losing access feels like grief. None of this proves harm on its own. Early research on cognitive offloading — the well-established idea that we hand mental tasks like memory or math to outside tools — suggests repeated chatbot use might dull the very skills it replaces; some studies have found reduced brain activity or weaker critical thinking tied to AI use, though the science is nowhere near settled.

What makes this worth watching isn't Green individually — his research-aid use doesn't look inherently harmful despite the outrage. It's the scale. OpenAI alone reported over 900 million weekly active users this year, and even a small slice of unhealthy use across a user base that size adds up to millions of people. Search engines and social media both took years before anyone properly understood what they were doing to memory, attention, and wellbeing. AI is following the same slow, uncomfortable timeline, and Green may just be one of the first well-known people willing to say the quiet part out loud.