A new study asked 1,058 youth and young adults about their use of large language models (LLMs like ChatGPT) for mental health advice. In total, 13.1% had asked AI for mental health advice, but this number was higher at 22.2% for young adults (18-21 years old).
Almost all of those surveyed felt that they received helpful advice from the chatbot (92.7%), although that number was lower for Black participants, indicating that LLMs may be reinforcing racial bias.
“High use rates likely reflect the low cost, immediacy, and perceived privacy of AI-based advice, particularly for youths unlikely to receive traditional counseling,” the researchers write.
They add that “engagement with generative AI raises concerns, especially for users with intensive clinical needs, given difficulties in establishing and using standardized benchmarks for evaluating AI-generated mental health advice and limited transparency about the datasets training these models. Furthermore, Black respondents reported lower perceived helpfulness, signaling potential cultural competency gaps.”
The study was conducted by researchers at RAND, Harvard, Brown, Mass General Brigham, and Boston Children’s Hospital, led by Ryan K. McBain and supervised by Jonathan Cantor. It was published as a research letter in JAMA Network Open.
An article in The Guardian on November 30 revealed that chatbots offer dangerous advice to those whose mental health is vulnerable.
The researchers contacted 2,125 youth and young adults (aged 12-21) via the internet through RAND’s American Life Panel and Ipsos’ KnowledgePanel. About half (1,058) responded; 37% were aged 18-21, 50.3% were female, and 51.3% were white.
The survey asked whether the participants used AI for “advice or help” with “feeling sad, angry, or nervous”—everyday language to ensure it was understandable even by the youngest participants.
In total, 13.1% had asked AI for mental health advice, but this number was higher at 22.2% for young adults (18-21 years old).
Of those who used AI for this purpose, 65.5% did so at least monthly (or more often) and 92.7% said the advice was at least somewhat helpful.
The researchers state that their results should be interpreted with caution. There was no data on the respondent’s actual psychiatric needs or whether they were receiving other forms of mental health intervention. Survey response bias may have influenced the results. The number of respondents who were 18-21 was small (147 participants).
The researchers speculate that youth may believe AI is more private than discussing these concerns with adults such as family, teachers, or therapists. Yet AI companies are recording these conversations, training their future AI models on them, and even reporting them to the police.
Under Trump’s guidance, the AI world is almost completely unregulated.
In terms of mental health bots, AI therapists are being marketed to users while encouraging delusional thinking and helping users plan suicide. They perpetuate stereotypes and reify inequalities, discriminating against marginalized people and even engaging in hate speech against those with psychiatric diagnoses.
Outside of the medical field, an expose revealed Meta’s AI policies, which specifically enabled chatbots to have sexual conversations with minors, promote misinformation, and more.
Google’s Character.AI has also been responsible for teen deaths, with a recent suicide spurred by the hypersexualized bot, which has also taught other teens how to self-harm and more.
****
McBain, R. K., Bozick, R., Diliberti, M., Zhang, L. A., Zhang, F., Burnett, A., . . . & Yu, H. (2025). Use of generative AI for mental health advice among US adolescents and young adults. JAMA Network Open, 8(11), e2542281. (Full text)
On the one hand, I’d honestly rather have people of any age turning to “AI” chatbots instead of therapists solely for the fact that “AI” is cheaper than therapy, if not free, and that it likely won’t turn out to be a controlling narcissist.
On the other hand, we need to stop calling these things AI. They’re word-salad machines. They don’t think, they don’t understand, they get facts wrong because that’s fundamentally not what they’re made to do. They put words together in the order that is most likely given the input. I do appreciate the nod toward racial bias, though, because that is sadly still true of even the word-salad spinners.
It’s really tragic that so many people, not just kids, turn to the word salad bar for emotional support, because it means they’re literally getting more interpersonal support from a non-human, unthinking noise generator than from the actual humans in their lives. THAT’S what people should be concerned with, not that it gives “dangerous advice”. Much of that advice, if they didn’t get it from word salad, they’d have searched somewhere else and found it.
The psychosis and “advice” that is just complete nonsense detached from reality… I hate to say that the problem there is that people think these things are intelligent. That they think they’re MORE intelligent than people and trust their “advice” and “opinions” like holy gospel. We need to be undermining people’s blind faith in word salad, not “improving” the salad bar to be more “accurate”.
I also have to wonder how much of this phenomenon is driven by illiteracy. When over half of americans read at or below a sixth-grade level, the machines that can generate any style or complexity of word salad must seem awfully intelligent.
Report comment
I have also found when testing these things out that they are reliant (of course) on things already written elsewhere. This tends to create a bias toward the status quo. One is not likely to be asked challenging questions or provided with a new perspective by AI. You’re going to get regurgitated psychology and psychiatry memes, disguised in dressed up “intelligent” language. No real intelligence involved. Though of course, you are right that at least the AI won’t develop hidden agendas and try to hurt you. But that’s a pretty low bar!
Report comment
Given how unhealthy it can be to turn to flesh-and-blood doctors for health advice, it’s hardly a surprise that AI (LLMs), drawing from a long history of Pharmafia medicine, ends up duplicating dubious science behind its organized crime bizzness for profit at people’s expense. It’s at the very least unadvisable to trust any product pushed upon us by powers that be, running rackets across a social system serving the same pathology of profit at large. Buyer beware all the more when it comes to technology that’s long been researched and developed in darkest depths of DARPA and all the other alphabet agencies of the military industrial complex and national security state for more perfect population control in the digital (c)age.
Just take precedent like television, or what once was called the boob tube. Here’s an earlier electronic addiction for social engineering that bears comparison with MK-Ultra mind control in more confined laboratories at the same time. Not unlike better living through chemistry by LSD, TV was a trip into hallucinatory states for the general public, brainwashed by propaganda better than any other opiate of the masses, and ensuring passive conformity to ‘the way things are’ by “amusing ourselves to death” (Neil Postman) in fantasy land. It’s just a matter of which method of lobotomy you choose, from the ‘alternatives’ chosen for you, and the normative ideal is both, and then some.
Digital tech makes the telly look like a prehistoric relic equivalent to painted pictures on cave walls. Its spells colonize consciousness at an entirely new lower level of dumbed-down identification with programmed conditioning undreamt of by all previous drug pushers. Its normative ideal, already in an advancing and accelerating process of implementation in the present, is to create biodigital convergence, our merging with the Machine to make human agency and autonomy, long reduced to resources for rulers of so-called civilization, obsolete if not extinct. Whatever conveniences we (double)think we gain by ‘smart’ technology are simply not worth such total alienation.
Every major postwar industry in ‘our’ nation has been the product of the military industrial complex (MIC), confiscating public wealth, like taxes as tributes to the Pentagod, for an increasingly fused (and fascist) corporate state engineering society for the profit and power of ruling elites. We here might recall Eisenhower’s famous farewell address to the nation, warning of the dangers of not only the MIC but science and technology subordinated to elite control. Given how his words were hypocritical after his own administration had served such purpose, as with rapid expansion of the nuclear arms race, we might also consider how ostensible concerns articulated for public consumption are crafted to cover up actual agenda antithetical to human welfare.
In relation to the report above, we read warnings of dangers to mental health, particularly of children, from the very participants helping to advance our continuing incorporation into the digital (c)age. The RAND Corporation has been one of the leading representatives of the MIC since WW2, responsible for some of the most Machiavellian methods of psychological as well as kinetic warfare. The academic industrial complex, above all elite institutions like the Ivy League universities, receives most of its R&D funding from (for) the military. Recalling Foucault’s focus on biopower, the medical industrial complex, now the largest sector of the national economy, has become indistinguishable enough from the military in pursuit of population control and biotech/bioweaponry that reference is now made to the MMIC.
It’s naive to believe negative effects of any technology are simply unknown to principal players in its R&D prior to publicized studies like that here. If anything, these negative effects, much like the ‘side’ effects of pharmaceutical poisons upon permanent ‘patient’ populations, are primary objectives of the ruling interests served by the R&D. As with product rollout of any brand-new, breakthrough treatment of ‘health care’, such publicized studies as here are part of the propaganda/psywar apparatus of mass communications selling controversy to information consumers, or junkies, obscuring real agenda coordinated beyond the limited hangouts, as the CIA says, scripted for us. Us, who according to experts can’t count on our common sense to know we’re selling our souls to screens to become ghosts in the machine. (Where are the luddites when you need them?)
What’s the difference between science and superstition when “the welfare of the people is always the alibi of tyrants” (Camus)? Why buy any advertising from liars pretending this mad American dream machine is not out to swallow us whole? What does ‘regulation’ mean besides foxes guarding the hen house under pretext of protecting us and not the predators among all the revolving doors of the corporate state? What can we expect if not more readymade solutions to manufactured problems, exploiting the old poster child appeal, more police state supervision for all infantilized subjects made to turn to our oppressors in exchange for our safe keeping?
Artificial intelligence is an inevitability of ‘progress’, imposed on us by our enlightened benefactors. We are permitted to argue, to no avail, its pros and cons on the way to its ultimate establishment, and final solutions. There’s the biggest lie of all, perhaps. Insofar as we worship class rule, mistaking the law of rule as the rule of law, the global chaos created by its crimes against humanity as order, we are prisoners to social insanity. A planetary panopticon is being prepared for us. If it’s not already past time to do so, we better wake up, and rise up to change this social system at its roots with radical egalitarian freedom from any god or master sacrificing us on its altar of rule.
For some sampling of the Frankenscience in the works:
Technocracy Videodrome
https://iceni.substack.com/p/technocracy-videodrome
For an example of the problem-solution and good-cop/bad-cop con game:
As Expected, a Hearing on Kids Online Safety Becomes a Blueprint for Digital ID
https://reclaimthenet.org/as-expected-a-hearing-on-kids-online-safety-becomes-a-blueprint-for-digital-id
Report comment
AI Is Destroying the University and Learning Itself
https://www.globalresearch.ca/ai-destroying-university-learning/5908071
Report comment
That was already happening. The argument could be made that AI is helping students get smarter, because otherwise they would be completely lost. It is not an issue of “students will work harder without AI” but more so “schools will keep lowering their standards to make it look like students are performing better”.
Report comment
“Almost all of those surveyed felt that they received helpful advice from the chatbot (92.7%), although that number was lower for Black participants, indicating that LLMs may be reinforcing racial bias.”
-It would be helpful if we could see what the actual numbers are, are they 92.6%? 10%? Also, is it really good science to infer that LLMs are reinforcing racial bias because Black participants had lower satisfaction rates? Wouldn’t that assumption be a racial bias?
Report comment