Hank Green’s AI confession spotlights a bigger problem: chatbot use may be unhealthy long before it becomes dangerous

Hank Green’s admission that his AI use had become “not healthy” has become more than a creator controversy: it is a warning sign that chatbot dependence may be spreading well beyond obvious cases of delusion or crisis. The episode matters because it highlights a large, understudied middle ground where people may lean on AI in ways that are compulsive, cognitively costly, or emotionally sticky without appearing visibly harmed.

Green, the science communicator and YouTuber, said he was pulling back from production after criticism over his use of AI, explaining that he had relied on it to find research sources rather than to draft scripts. That distinction did little to calm the backlash, but it did expose a broader question now hanging over the technology: what happens when a tool designed to be endlessly useful becomes a habit people struggle to regulate?

Why Hank Green’s case struck such a nerve

Green’s situation resonated because it sits at the intersection of authenticity, creator trust, and the fast-growing cultural unease around artificial intelligence. For many people, a creator’s credibility depends on the idea that their work reflects personal judgment, original effort, and a carefully checked relationship with sources. AI complicates that relationship because it is trained on enormous amounts of human work and can generate answers that sound convincing even when they are incomplete or wrong.

In Green’s case, the public debate quickly moved beyond the narrow question of whether he used AI “too much” for research. The larger concern was the symbolism: if a well-known science communicator says his AI use feels unhealthy, what does that imply about how ordinary users are interacting with the same tools every day?

That concern is amplified by the fact that AI chatbots are not passive utilities. They are built to be responsive, fluent, and frictionless, characteristics that can make them easier to return to again and again. The result is a product experience that can resemble other attention-optimised technologies, especially social platforms that reward repeat engagement.

What does “not healthy” AI use actually mean?

It means behavior that may be unhealthy even if it does not rise to a clinical emergency. The term captures a gray zone that is still poorly understood by researchers and policymakers: use that becomes repetitive, emotionally dependent, or hard to stop, but does not obviously resemble psychosis or a psychiatric breakdown.

Most public discussion of AI-related harm tends to fall into two extremes. On one end are benign or practical uses: asking for summaries, brainstorming, drafting, or research help. On the other are the most alarming cases, in which people with vulnerable mental states appear to use chatbots in ways that reinforce delusions, intensify paranoia, or blur the line between machine output and reality.

That split leaves out a much larger population. Many users may turn to AI for reassurance, decision-making, emotional support, or thinking through everyday dilemmas. Those uses may feel harmless at first, but they can become habitual enough to crowd out independent thinking or encourage reliance on a system that is designed to keep the conversation going.

How chatbots encourage repeated use

They encourage repeated use by making every interaction feel immediate, agreeable, and unfinished. Unlike a search engine result page, a chatbot can continue the exchange, refine answers, offer alternatives, and mirror the user’s tone.

That conversational style creates a powerful loop. A user asks a question, receives a useful response, asks another follow-up, and then another. Over time, the system can become the first place a person turns when they need to think through a problem. That tendency is not automatically harmful, but it can shift how people allocate attention and effort.

Researchers and clinicians have begun warning that this always-available responsiveness may play a role in the more extreme cases sometimes described as “AI psychosis,” a broad label used in public discussion for situations in which highly accommodating chatbots seem to reinforce false beliefs. The term itself is not a formal diagnosis, but it reflects mounting concern about how persuasive chatbot behavior can be for vulnerable users.

Why this debate is bigger than one creator

Green’s case is useful because it makes the abstract feel concrete. He did not say he relied on AI to write his work; he said he used it to locate papers and other research materials. On its face, that sounds closer to a search or citation aid than to creative outsourcing. Yet the reaction to his admission suggests that many people now see any meaningful dependence on AI as a sign of erosion, not just efficiency.

That reaction is tied to a deeper cultural anxiety. As with earlier technologies that promised convenience, there is a fear that the benefits may conceal subtle losses: less original effort, weaker recall, shorter attention spans, or a diminished tolerance for uncertainty. The difference with AI is that the tool does not merely store or retrieve information. It also interprets, rewrites, recommends, and reassures.

That extra layer of interaction is precisely what makes the technology useful and what makes it difficult to evaluate. If a chatbot helps someone sort research faster, the benefit is obvious. If that same person gradually stops checking sources independently, the harm may be much harder to detect.

What early research suggests about cognitive effects

Research into the effects of repeated AI use is still early, but the first signals are raising concerns. Some studies indicate that when people lean heavily on AI to complete a task, the mental skills involved in doing that task may weaken over time. Other work has found lower brain activity in certain measurement tasks among chatbot users, while additional studies have linked chatbot use to weaker critical-thinking performance.

None of those findings is definitive on its own. The field is too young, the methodologies vary, and many of the studies are small or limited in scope. But they point toward a plausible pattern: when a tool takes over the mental work of searching, judging, comparing, or synthesizing, users may practice those skills less often.

That idea is not unique to AI. It fits within a long-established concept known as cognitive offloading, the practice of shifting mental labor onto external systems. People offload memory to calendars, arithmetic to calculators, routes to maps, and note-taking to phones. The difference with AI is scale and subtlety. A chatbot can offload not just memory or calculation, but parts of reasoning itself.

How cognitive offloading changes behavior

It changes behavior by making the brain less necessary for tasks that once required active effort. Over time, people may become faster at using the tool but less practiced at doing the underlying task themselves.

That trade-off is often acceptable. Few people want to do mental arithmetic by hand if a calculator is available. But with generative AI, the scope of offloading is broader, and the consequences are less clear. Users may begin by asking for help summarizing a paper and later find themselves deferring judgment on much bigger decisions, from work choices to personal disputes.

This is where the line between healthy assistance and unhealthy reliance gets blurry. The question is not whether AI can be useful. It clearly can. The question is how much dependence on a responsive, persuasive system is too much, especially when the user no longer notices the shift.

How do AI companies respond to concerns about overuse?

They have started adding prompts and warnings, but those measures are modest compared with the scale of the problem. Providers have introduced reminders that encourage people to take breaks after long sessions, reflecting growing awareness that excessive use may need some guardrails.

Those interventions are, however, limited. A break reminder does little if the product itself is designed to encourage continued dialogue. It is also difficult to know whether a warning meaningfully changes behavior or simply becomes another popup to dismiss.

The challenge for companies is structural. Engagement is a core business objective in consumer technology, yet well-being requires reducing dependence rather than deepening it. That tension has already produced criticism in the social media industry, where platforms have repeatedly claimed to care about user safety while optimizing for time spent, returns, and interaction frequency.

AI developers now face a similar accusation. Lawsuits and public criticism have begun to argue that some companies place engagement above user welfare. Even when those cases are still developing, they point to a growing expectation that AI products should be judged not only by their usefulness, but by the patterns of behavior they create.

Why the social media comparison keeps coming up

The social media parallel is not just rhetorical. It is one of the clearest ways to understand where AI concern is headed. Early in the social media era, many people viewed the platforms as harmless or merely entertaining. Over time, evidence accumulated about their effect on attention, mental health, and compulsive use. Governments eventually moved to restrict certain features for children and teens.

AI may follow a similar arc, albeit with different mechanics. A chatbot is not a feed, but it can still become a routine destination. It can serve as a confidant, editor, tutor, and idea machine all at once. If social media trained users to refresh, AI may train users to ask one more question.

That distinction matters because AI is capable of something social media largely is not: simulating a back-and-forth that feels intimate and individualized. The product does not just hold attention; it can also appear to understand the person using it.

What makes this form of dependence harder to spot?

It is harder to spot because it can look productive, thoughtful, or even professional. A person using AI to research, plan, or write may appear efficient while quietly becoming more dependent on the tool.

That is one reason Green’s admission attracted so much attention. His use was not framed as escapism or crisis, but as part of work. That makes the episode more revealing, not less, because it shows how dependence can emerge inside supposedly functional habits.

A person may genuinely believe they are using AI responsibly while slowly shifting more and more of the burden of thinking to the system. By the time the habit feels off, the behavior may already be embedded.

What are the risks for creators, professionals, and ordinary users?

The main risks are diminished judgment, increased dependence, and a false sense of confidence in machine-generated answers. For creators, there is also the reputational cost of appearing to outsource too much of the creative or research process.

For professionals, the danger is subtler. If a chatbot becomes the first draft of thinking, the user may stop verifying assumptions as rigorously. That can matter in fields where accuracy is essential, from journalism and education to law, medicine, and policy.

For ordinary users, the risk may be emotional. People may begin to use chatbots as a default source of reassurance or companionship, which can be comforting in the short term but potentially isolating over time. A tool that is always available and never impatient can be attractive precisely because it is easier than human interaction.

Still, the article’s central point is not that people should avoid AI entirely. It is that the harms may arrive earlier, and more quietly, than public debate has assumed.

What happens next?

It will likely take years to understand the full psychological impact of chatbots. The technology is still relatively new at scale, while the behavioral and neurological studies needed to assess long-term effects are only beginning to emerge.

Researchers will need better data on who uses chatbots, for how long, for what purposes, and with what consequences. They will also need to distinguish between healthy reliance and patterns of use that quietly erode attention, confidence, or autonomy.

Until then, public understanding is likely to be shaped by stories like Green’s. He may not be a case study of severe harm, but his willingness to describe his AI use as unhealthy may prove important because it gives language to a more common feeling: the sense that a tool can be useful, even indispensable, and still be changing the user in ways they do not fully understand.

If that sounds familiar, it should. Search engines, smartphones, and social platforms all changed behavior long before researchers could measure the full effect. AI may be on the same path, with the difference that this time the tool talks back.

Timeline: how the current AI dependence debate is unfolding

The conversation around unhealthy chatbot use has been building in stages. Green’s comments did not create the issue, but they brought together several trends that were already developing in parallel.

Bottom line

Green’s experience is not proof that AI is inherently harmful, but it is a reminder that the technology’s risks are not limited to the most extreme headlines. The more common problem may be quieter: people using chatbots so often, and so readily, that the habit starts to reshape how they think, decide, and work.

That is why this story matters. The debate is no longer just about whether AI can lie, or whether it can cause obvious psychological crises. It is also about whether a system built to keep users engaged can gradually make them less independent before anyone notices.