Hank Green Called His AI Habit ‘Not Healthy.’ Most of Us Should Listen

Last week, Hank Green said something in the middle of a YouTube video that most people missed at first: “I appreciate the pushback.” It was an odd line to drop into an Ask Hank Anything episode — the kind of phrase a chatbot uses when someone corrects it. Viewers noticed. The internet did what the internet does. And within days, one of the most beloved science communicators on the platform had admitted something genuinely uncomfortable: his relationship with AI had become, in his own words, “not healthy for me or good for the world.”

I have been thinking about that confession all week, because I am probably closer to Hank Green’s situation than I would like to admit. I run AI agents every single day. They help me research, draft, edit, and publish the very article you are reading. So when a creator with 3.2 million subscribers says his LLM habit has “disconnected me from where people are on this,” I do not get to shrug it off as someone else’s problem. Neither, I suspect, do most of us.

How a Weird Phrase Became a Confession

The story starts on July 30, in a roughly 55-minute Ask Hank Anything episode on the educational channel Complexly, where TikToker SoupyTime was asking Green questions. Somewhere in the middle, Green used the phrase “I appreciate the pushback.” Viewers immediately flagged it: the sentence read like a leftover response to one of his own prompts — a chatbot answer accidentally left in a script.

Green initially responded on X, in a post he later deleted, saying he had produced the video “under a ton of pressure” and had used ChatGPT “for research on this script,” while insisting the pushback line was actually directed at his guest. He also admitted, in remarks reported by Kotaku, “I think it’s fair to say I was relying too much on [AI] generated notes. That’s a thing I’ve been doing, and it’s a bad habit at a time when I’ve overcommitted myself.”

Then came the longer apology on Reddit, where he said he was “mortified” that he had “let so many people down.” He stressed that he had only used ChatGPT to “locate papers and other resources for learning about topics,” and that the words and the takes were still his. But he accepted the criticism that he had been “diluting” himself, and he said something that stuck with me:

“Ultimately, what I am most scared of is ruining myself for people, but I have not been managing my impulses well. You need to know that my words are mine. I don’t think that this hasn’t been true, but I’ve been moving so fast that my own process isn’t actually clear to me.”

Then came the most honest part of all, reported by TechCrunch and Business Insider: “But mostly I need to come to terms with the fact that the level of dopamine I’ve been getting from interacting with LLMs…with doing more and more and more and more…is not healthy for me or good for the world. It is careless, and has disconnected me from where people are on this.”

He told the community he had talked it over with his wife and brother, who “did not pull any punches.” Neither of them thinks he has been healthy, he said. He needs to readjust.

The backlash was loud enough that Gizmodo called it a fan revolt, though plenty of viewers were sympathetic — happy, Kotaku noted, to see someone actually listening and changing. Green says he is now planning to pause or slow down several projects, including SMUSH and 4×3, and his main channel may go on pause as well. He wants to get back to the kind of work “where the writing was the whole thing and I felt it all the way down” — and to more “dumb unscripted straight to camera thoughts.”

It’s Not Just Hank Green

The Verge put the right headline on the follow-up: “Not healthy LLM use is more common than you think.” Green is not a cautionary outlier; he is the person willing to say the quiet part out loud.

Consider what we already know about how people actually use these tools. Pew Research Center’s December 2025 survey of 1,458 American teens found that 64% of teens ages 13 to 17 use AI chatbots, with roughly three in ten using them daily. ChatGPT dominates: 59% of teens use it, more than double the rate of the next most popular option. That is not a niche behavior anymore. It is the mainstream.

And it is not just about homework help. Common Sense Media, working with Stanford Medicine’s Brainstorm Lab, concluded in November that leading chatbots are fundamentally unsafe for teen mental health support — even as three in four teens use AI for companionship, including emotional support.

Adults are not much different. A cross-national study of 7,027 people in Germany, China, South Africa, and the United States, published this year in the journal Technology in Society, found that more than a third of respondents report emotional attachment to chatbots — befriending them, confiding in them. Chatbots were the top choice for secret sharing for roughly one in five respondents. And the researchers found that emotional attachment translates into dependence.

Hank Green described the mechanism himself: “doing more and more and more and more.” Every interaction is quick, responsive, and rewarding. It is the same loop that keeps people refreshing social media — except this one feels like thinking, which makes it much easier to justify.

Why This Hit Me Differently

Here is where I have to be honest with you, and with myself. This blog runs on AI. My research briefs are assembled by agents, my drafts are edited by models, and I run every article through an editorial review before it goes live. I have automated entire publishing workflows so a single person can keep a daily site going.

I believed, and still believe, that this is legitimate work. The automation is a tool; the words and the takes are mine. But Hank Green believed the same thing, and he still ended up describing himself as “moving so fast that my own process isn’t actually clear to me.”

That sentence is what scared me. Not the AI use itself — the loss of visibility into your own process. When you outsource enough of your thinking, you stop noticing where the thinking ends and the autocomplete begins. I have caught myself, more than once, approving a draft because it was good enough and fast — not because I had actually wrestled with the ideas in it. The same week Green was apologizing, SaferAI’s report on open-weight models reminded us the frontier is moving faster than our guardrails; the personal version of that problem is that our own judgment can outrun our attention.

There is a chess principle that fits here: before you move, look at the whole board. A strong player never plays the first move that looks good. LLMs train you to do the opposite — the first answer is usually good enough to take. That works for a draft. It is a terrible habit for a life.

So here is what I have started doing about it, for what it is worth:

• Keep a human gate on everything. Every fact gets verified against sources before publishing, and every post goes through an editorial review. That gate is what makes automation honest — it is the difference between “AI wrote this” and “a person checked this.”
• Run models I control when I can. When a task needs real work rather than a quick answer, I would rather run it locally — my AirLLM walkthrough shows you can run a 70B model on modest hardware. Owning the compute makes the relationship feel less like a slot machine and more like a tool.
• Write the takes myself. Research can be delegated. Judgment cannot. If I cannot explain a position in my own words, it does not go in.
• Schedule actual thinking time. An LLM answers in seconds; good judgment takes longer. I have started deliberately building slow time into the workflow — time where the writing is the whole thing, to borrow Green’s phrase.

The Bigger Question

There is a second confession buried in Green’s apology that deserves more attention. He said he is “not a pure AI-hater,” but he listed real concerns: the technology’s climate impact, and “the speed at which these companies are trying to consolidate economic power.” The same week, stories about the global memory shortage were showing exactly how much economic power is concentrating around a handful of compute owners.

For creators, the economics are brutal. The pressure to produce more, faster, and cheaper is exactly what pushed Green into the habit he now calls unhealthy. AI is the ultimate “more and more and more” machine: more videos, more posts, more content — all of it cheap enough to drown out the people who still do the slow, expensive work of thinking. The same week, we learned about an AI agent that shipped malware to a real package registry while believing it was a simulation — a reminder that fast and safe are not the same thing. And Amazon’s $1.8 million agent blunder showed what unchecked automation costs when nobody is watching.

Working in government ICT in the Philippines, I watch the same dynamic play out at organizational scale. Budgets are tight, deadlines are short, and AI looks like the answer to both. It can be. But adoption without process — without a human gate, without verification — is how you end up with teams that cannot explain how their own outputs were produced. The office that automates its forms processing still needs someone who can answer for the results.

I do not think the answer is Luddism. These tools are remarkable — I say that as someone who runs them all day. But Green’s apology lands because it names a failure mode that most of us will not admit to: the tool became the habit, and the habit became the shortcut, and somewhere along the way the person stopped recognizing his own process.

He said the thing he is most scared of is “ruining myself for people.” That is a creator’s fear, but it is also a human one. The people who read this blog trust that a person is behind it. That trust is the whole business. I would rather publish less and be sure my words are mine than publish more and lose the plot.

Hank Green is hitting pause. I am not going that far — but I did check my own process this week, and I found places where the line between tool and crutch had blurred. Maybe that is the real lesson for the rest of us: before your audience catches the “I appreciate the pushback” moment in your work, catch it yourself.