Every self-help book is making roughly the same pitch: here’s a framework, follow it, and you’ll build the discipline or the mindset you’re missing. That pitch now has competition that costs nothing, fits in a pocket, and never judges you for opening it at 2 a.m. AI Chatbot Use and Disclosure for Mental Health Among US Adolescents and Young Adults, published in JAMA Pediatrics on June 1, 2026 by researchers at RAND, found that 19.2% of Americans ages 12 to 21 (roughly 8.2 million people) used an AI chatbot for mental health advice in 2025. Nearly two-thirds never told a single person. Weeks later, the American Psychological Association’s 2026 Chatbots and Mental Health Survey of more than 1,200 licensed psychologists put a number on what those conversations are actually about: 34% of patients are using AI specifically for self-discipline, affirmations, or behavioral reminders. Read that list again. That’s the self-help book’s whole job description, minus the book, the accountability, and anything resembling an evidence base.

Quick Verdict

Question	Answer
How many young people used AI for mental health advice in 2025?	19.2% of US 12-21 year olds (about 8.2 million people)
How many told someone about it?	37% (63% disclosed to no one)
How many psychologists have patients bringing AI into session?	77%
How many use it for self-discipline, affirmations, or reminders?	34%
Signs of dependency reported by psychologists	36%
Patients using AI to self-diagnose	39%
Patients developing delusional beliefs after chatbot use	15%
Patients forming an “intimate or relationship-like” bond with a chatbot	13%

Bottom line: The audience for self-discipline advice hasn’t shrunk. It’s just moved to a product with none of the guardrails a published book has, and most of that movement is happening where nobody (a parent, a therapist) can see it.

What the RAND Study Actually Found

The RAND survey behind the JAMA Pediatrics paper polled 1,009 adolescents and young adults and weighted the results to represent roughly 42.8 million people nationally. The 19.2% figure for 2025 is up from 13.1% the year before, close to a 50% jump in twelve months and a fast adoption curve for anything, let alone something this personal. Among the people who did use a chatbot for mental health advice, 42.8% did it at least monthly, and 91.7% rated the advice as somewhat or very helpful. Usefulness isn’t the concerning part of that data. The concerning part is what comes next: 63% of users disclosed the practice to no one. Not a parent. Not a therapist. Nobody.

Compare that to a book. A teenager reading a self-help book leaves a trail — it’s on a nightstand, in a backpack, checked out from a library, gifted by a parent who read the back cover first. A chatbot conversation about self-discipline or anxiety leaves nothing. No spine on a shelf for anyone to notice, question, or follow up on.

What Psychologists Are Watching Happen in Session

The APA’s 2026 Chatbots and Mental Health Survey surveyed licensed psychologists directly rather than patients, and the picture from that side of the room backs it up. Seventy-seven percent said their patients bring up AI use unprompted. Beyond the 34% using it for self-discipline, affirmations, or behavioral reminders, 33% reported patients using AI to assist with treatment and 35% said patients treat it as an additional mental health professional, a stand-in for a second opinion, or sometimes the only opinion.

How Many Teens Use AI Chatbots for Mental Health Advice?

1. 19.2% of US 12-21 year olds used an AI chatbot for mental health advice in 2025, up from 13.1% in 2024.
2. That’s an estimated 8.2 million young people nationally.
3. 63% told no one they were doing it.
4. 42.8% of users checked in with a chatbot at least once a month.
5. 91.7% rated the advice they got as somewhat or very helpful.
6. Usage is now close to the 19.8% who reported getting help from an actual mental health professional in the same survey.

That last point is the one worth sitting with. Chatbot use for mental health advice isn’t a fringe habit anymore. It’s running roughly even with professional counseling among the same age group.

The Self-Help Book’s Pitch, Minus the Book

Self-discipline, affirmations, behavioral reminders: that’s not a random cluster of use cases. It’s the back-cover copy of half the self-help section. We’ve spent time on this site picking apart what actually works in that genre versus what’s packaging: the habit-cue mechanics buried in Atomic Habits once you strip the anecdotes, and the all-or-nothing discipline sprint that TikTok keeps rebranding every fall. A chatbot doesn’t sell you a framework. It generates one, on demand, tailored to exactly what you typed, with no editor, no publisher fact-check, and (this is the part that should give you pause) no research behind the specific advice it just gave you at 11:47 p.m.

There’s a specific overlap worth naming directly: “behavioral reminders” is one of the three uses the APA survey measured, and it’s functionally identical to the second-brain, offload-everything advice that fills the productivity shelf. We covered a study earlier this month showing that people who lean on reminders for a task end up worse at remembering it once the reminder disappears. Not neutral: worse than people who never had a reminder at all. If that’s true of a to-do app notification, there’s no reason to assume a chatbot nudging you toward a habit works differently. The mechanism (effortless prompting instead of effortful recall) is the same regardless of whether the reminder comes from an app or a conversation.

Is Using an AI Chatbot for Self-Help Safe?

Sometimes, for low-stakes prompting like a reminder or a reframe. Not reliably, for anything clinical. The APA survey found real risk signals alongside the helpfulness: 36% of psychologists reported patients showing signs of dependency on a chatbot, 39% reported patients using AI to self-diagnose, 15% reported patients developing delusional beliefs after chatbot conversations, and 13% reported patients forming an “intimate or relationship-like” connection with one. None of that shows up as a risk on a self-help book’s back cover. A book can’t argue with you or agree with everything you say. It’s not awake at 3 a.m. when you’re not thinking clearly.

Those four numbers matter more together than separately. A book that’s wrong sits there being wrong until someone corrects it. A chatbot that’s wrong keeps talking, keeps agreeing, and (per the self-diagnosis and dependency figures) keeps getting asked. We’ve written before about self-help products that promise change with no measurable follow-through. At least a book’s failure mode is boring: you read it, nothing changes, you move on. A chatbot’s failure mode, per this data, can be a patient walking into a session already convinced of a self-diagnosis, or already emotionally attached to the thing that got them there.

The Secrecy Problem

Sixty-three percent telling no one isn’t a footnote. It’s arguably the headline. Self-help books, for all their flaws, are a public act — someone sees you reading, sees the title, sometimes asks about it. That visibility is a weak but real check: it invites a conversation, occasionally an intervention. A chatbot conversation about feeling anxious or lonely has none of that. It’s the most private version of seeking help that’s ever existed, and privacy cuts both ways. It removes stigma. It also removes anyone who might notice a 12-year-old is struggling and it’s gone on for months.

We’ve looked at loneliness and isolation as genuinely different problems that require different fixes: isolation is structural, loneliness is a felt gap, and books address the second one at best. A chatbot that’s always available and never disclosed to anyone is a plausible way to paper over both without actually fixing either. It feels like connection. It isn’t one, in the way a phone call or a person sitting across from you is one.

What a Book Still Does That a Chatbot Doesn’t

None of these are trivial. A published book went through an editor and, in the better cases, cites actual research. You can check the claim, even if the citation turns out to be cherry-picked; a chatbot’s advice is generated in the moment and disappears with the conversation, with nothing left to audit. A book also doesn’t adapt to tell you what you want to hear. It says the same thing to everyone who opens it, which is a limitation and also a safeguard: the sycophancy risk that shows up in the dependency and self-diagnosis numbers above doesn’t really exist on a printed page. And reading a book is slow enough to create a gap between the urge for help and the response, the gap where a person usually decides whether they need more than a paragraph of encouragement. A chatbot closes that gap instantly, which is exactly why it’s useful for a quick reframe and exactly why it’s a bad substitute for anything that needs slower, professional judgment.

When the Chatbot Makes Sense, and When It Doesn’t

• Fine for a quick nudge. “Help me reframe this one thought” or “remind me why I set this goal”: low-stakes, in-the-moment prompting is close to what a good affirmation exercise in a workbook does anyway.
• Not fine as a stand-in for therapy. The self-diagnosis and delusional-belief numbers above are exactly the failure modes you’d expect from a tool that has no training, no license, and no way to know when it’s out of its depth.
• Not fine when nobody else knows. If chatbot use for mental health support has been going on for months and it’s a secret, that secrecy is itself the flag, not the AI use specifically.
• Worth naming out loud to a real person. The disclosure gap (63% telling no one) is the single most fixable number in either study. It doesn’t require banning anything. It requires one conversation.

The Bottom Line

The self-help industry has spent decades selling frameworks for discipline, affirmation, and behavioral change. In 2026, a meaningful share of the people who’d have bought that book are instead typing the same request into a chatbot, for free, in private, and without anyone checking whether the advice they got back was any good. The demand didn’t go away. It moved to a format with faster response times and weaker guardrails. A book that’s honest about its own limits, with real research behind its claims, is still the safer bet for anything that isn’t a quick reframe. The data says fewer people are reaching for that book, and most of them aren’t telling anyone what they’re reaching for instead.

---

Sources: AI Chatbot Use and Disclosure for Mental Health Among US Adolescents and Young Adults, JAMA Pediatrics, June 1, 2026; RAND press release; APA 2026 Chatbots and Mental Health Survey and APA press release. Related reading: Atomic Habits Review, Does ‘The Great Lock In’ Trend Actually Work?, Your Reminder App Is Quietly Wrecking Your Memory, Self-Help Products Don’t Change Your Personality: Study, and Can Personality Books Actually Cure Loneliness?.