Hank Green, the YouTuber and science communicator, is stepping back from production after describing his own reliance on AI chatbots as 'not healthy.' Green said he used the tools to hunt down research sources rather than to write scripts, but the admission has landed in the middle of a much larger question: how many of the 900 million people OpenAI says now use ChatGPT every week are quietly in the same place he is.
OpenAI disclosed the 900 million weekly active user figure earlier this year, and that number covers only one provider. Comprehensive data across every chatbot on the market is hard to come by, but the scale means even a small percentage of unhealthy use translates into millions of people. Green may be one of the first high-profile creators to publicly articulate the feeling, but he is almost certainly not the first to have it.
The backlash against Green centered on the tension between a brand built on authenticity and a technology trained on other people's uncompensated work, one that also fabricates plausible-sounding citations. He apologized. The louder signal, though, was the personal one — a working professional describing a habit he could not fully control around a tool most of his critics also use daily.
Key facts
- 01Hank Green is stepping back from YouTube production after describing his own AI chatbot use as 'not healthy.'
- 02OpenAI said this year it has more than 900 million weekly active users of ChatGPT.
- 03Green said he used AI to find research sources, not to write scripts, but still called the reliance unhealthy.
- 04Early studies link repeated chatbot use to reduced brain activity on certain tasks and weaker critical thinking skills.
- 05Lawsuits have begun alleging that AI providers prioritize engagement over user well-being, echoing social media claims.
Public discussion of AI's psychological effects tends to sit at two poles. On one end is casual, apparently benign use. On the other are the extreme cases involving delusion, psychosis, and psychiatric care that have surfaced in reporting and litigation over the past two years. The space in between — compulsive, dependent, or otherwise unhealthy use that never tips into visible crisis — is where most people actually live, and it is barely studied.
That gap is a design outcome, not an accident. AI chatbots are built to keep chatting. Their conversational structure, agreeable tone, and open-ended prompts are engagement mechanics, the same category of design choices that made social media feeds sticky. Providers have begun adding gentle nudges, such as prompts to take breaks after long sessions, but the underlying incentive to maximize time-on-app has not changed.
Lawsuits filed against chatbot providers have started making the engagement-over-well-being argument explicit, borrowing the legal template that social media companies have been fighting for the past decade. The claim is that highly agreeable models reinforce whatever the user brings to them, and that this reinforcement — helpful in the median case — becomes harmful at the tails.
Green's specific use case, treating a chatbot as a research assistant to surface papers and background material, is not obviously in the harmful tail. Yet the research on cognitive offloading is not encouraging even for the mundane cases. Early studies suggest repeated chatbot use weakens the skills the tool replaces, that users show measurably less brain activity on certain tasks, and that heavier reliance correlates with reduced critical thinking scores. None of it is conclusive. All of it rhymes with the arc of prior technologies.
Search engines rewired how people remember information. Social media rewired attention and, according to a growing body of evidence, adolescent mental health. Governments in multiple countries are now imposing age restrictions on social platforms, roughly fifteen years after those platforms hit mass adoption. AI is on a faster ramp. ChatGPT went from launch to 900 million weekly users in under four years, a curve that leaves regulators, researchers, and users themselves with almost no time to develop the vocabulary for what is happening.
There is also a category of use that does not fit the research-assistant frame at all. Reports have grown of people turning to chatbots to think through personal problems, make decisions, or seek reassurance, and of users forming emotional bonds strong enough to produce grief when a model is deprecated or its personality is changed. These behaviors are not psychiatric emergencies. They are also not nothing.
The counterweight is that plenty of chatbot use is genuinely productive. Coders ship faster. Non-native speakers write more confidently. People with learning differences get patient, on-demand tutoring. Green himself defended the underlying task — locating sources — as legitimate. The question is not whether AI tools are useful. It is whether the interface pattern they use is compatible with the way humans regulate their own attention, and the honest answer is that no one knows yet.
For the AI industry, Green's admission is a preview of a coming reputational problem that funding rounds and benchmark scores will not solve. The social media playbook — deny, delay, then accept regulation a decade late — is available and, based on the current legal posture of several providers, is being followed. A different playbook exists: publish usage data, fund independent research on compulsive use, and design defaults that assume the median user needs friction rather than more engagement. Whichever path providers choose will shape the regulatory response more than any model release.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
