How Are Young People Using AI Chatbots for Mental Health? AI and Mental Health – Ep. 7

Conversation with Shums Alikhan, lead researcher of a new study on how young people are using AI for mental health support.

Transcript

Shums Alikhan 0:00

We found that 12% of that sample was specifically turning to generative AI to discuss their mental health. A third of these youth are using it as a substitute to getting care, meaning this is sort of their endpoint, meaning they are not following up with professional support services and actual human support. Yes, it's not a mental health tool per se, but that's what many youth are using it for. So it may not have been intentionally created for that, but that's what it has become.

Stephanie Hepburn 0:38

This is Crisis Talk. I'm your host, Stephanie Hepburn. In AI News, the AI company Anthropic is suing the U.S. government for banning all federal use of its large language model Claude. According to a Time article published on March 11th, the Pentagon had attempted to renegotiate Anthropic's government contracts to allow all lawful use of a classified version of Claude. In response, Anthropic CEO Dario Amade requested two specific guardrails that the large language model not be used for mass surveillance of Americans, nor in fully autonomous weapons systems. The government didn't agree, calling Claude a supply chain risk and prohibiting federal use of the AI model. Anthropic is now suing the government to stop the ban, and on March 26, a federal judge in California temporarily blocked the government from blacklisting Anthropic and labeling it a supply chain risk to national security. This week I'm speaking with Shums Alikon, a senior manager at Cirgo Health. She's the lead of a new study on how young people are using AI for mental health support, which found that 12% of the youth surveyed are turning to AI chatbots to discuss their mental health concerns, with 33% turning to large language models instead of professional support. A quick conflict of interest disclaimer. I discovered my books are part of a pirated book database called LibGen used to train AI chatbots. I'm part of a class action lawsuit against Anthropic. Let's jump in.

Shums Alikhan 2:06

I'm Shums Ali Khan. I'm a senior manager at Cirgo Health. Cirgo is a public benefit corporation, and our goal is to really find out the why behind human behavior. I lead our youth mental health portfolio and our youth mental health work, which is what we're going to be talking about today.

Stephanie Hepburn 2:26

Tell me about the Youth Mental Health Tracker.

Shums Alikhan 2:28

The Youth Mental Health Tracker, which is funded by Pivotal and Paramount Studios, was launched in December of 2024. And really the goal is to have it be the most comprehensive and actionable mental health data platform that allows policymakers, advocates, researchers, anyone in the youth mental health ecosystem to get a holistic perspective on the lived experiences of youth. And so we dive into how youth are feeling. You know, what is their well-being looking like? What does it mean for youth to thrive? And we do a deep dive by looking at things such as social connectedness, isolation, belonging, and really getting into those drivers of well-being. We just did a survey last fall. It was actually fielded in October and November of 2025, where we specifically asked youth ages 13 to 24 on how they are interacting with AI, what does AI use look like to them? How is it making them feel? And we also asked them broadly, like what is their mental health status, so we can get a better understanding on how they're engaging with AI and how it's impacting their daily life. Shums, what did you find? We found that 12% of that sample was specifically turning to generative AI to discuss their mental health concerns. Now, from this 12%, we also went a little bit deeper to find that 69% of those youth are relying on general purpose tools that are not designed for mental health support. So ChatGPT, Claude, for example, Gemini. A small percentage, 21%, is using both general purpose tools and a mix of mental health-specific tools. So kind of a mix there, but the majority is turning to just general purpose tools that are not intended for mental health purposes. We looked at the data a step further as well, and we found that of those that are turning to generative AI for mental health purposes, 41% did not report being encouraged to get professional help. So ChatGPT, Claude, Gemini, you know, whichever generative AI tool they use, it didn't refer them to professional support or human support. It didn't encourage them to get professional help, crisis services, et cetera. A third of these youth are using it as a substitute to getting care, meaning this is sort of their endpoint, meaning they are not following up with professional support services and actual human support. Whereas two-thirds of youth reported that they may use it as a complement to getting care.

Stephanie Hepburn 5:21

How did they feel about it using the large language models to discuss their mental health challenges or a personal challenge that they're experiencing? How did they feel during and after? Is that information you collected as well?

Shums Alikhan 5:36

Yeah, we definitely did. And it's really interesting because it was mixed. So some youth said that it was great to have someone to listen to. For example, one quote that really jumps to mind I felt more disclosure. Honestly, it felt better than getting advice. Um, it was someone to talk to that would help me work through problems in an unbiased way, for example. So there was some positive feedback. But then on the other hand, some youth also said it provided temporary relief. So one individual said it was like putting a band-aid on a giant wound or a giant gash. Another individual said, I felt the same. I got that temporary relief, but in the long run, there wasn't really much added benefit, but it does provide that temporary relief.

Stephanie Hepburn 6:28

I'm curious if there are differences between the individuals who were using it in a supplementary way versus those who are using it in lieu of a mental health provider.

Shums Alikhan 6:38

Youth with mental health struggles who turned to Gen AI, you know, as a substitute to care, they were two and a half times more likely to report not knowing it was possible to get professional help in the first place. Another significant portion also mentioned that they had a lack of support from their parents and their caregivers. Other barriers that were mentioned were transportation, cost, insurance. So they're turning to generative AI as a substitute because they feel as if there is no other option.

Stephanie Hepburn 7:10

Were you able to connect people to care or let the youth know what options were available to them? I know that's different than trying to collect data, but I'm curious if you use that also as an opportunity to provide information and resources for people who might be struggling.

Shums Alikhan 7:27

Yeah, that's a great question. You know, ultimately our reports and our tool are designed for policymakers, advocacy organizations, community-based organizations, other mental health experts working in the field. Parents and educators, of course, fall into that as well. So we did work with partners to develop recommendations on what the greater ecosystem can do to help youth get the help that they need because they are turning to generative AI. And one of the recommendations really falls around developing digital literacy programs to help youth understand, you know, when is it important to turn to professional support and human support versus generative AI?

Stephanie Hepburn 8:11

So I watched one of the interviews. The interviewee was a 19-year-old named Sydney, and she said, in order to help teens, we need teens. She also said if everybody just found one person, one trusted person that they could talk to, that would go so far. Did you find her quotes reflected in the overall survey?

Shums Alikhan 8:28

Yeah, we found that yes, peers are very important to youth, of course, at that age specifically. But their number one source of getting information, it seems to be parents. Over 80% of youth said that we still turn to our parents as our number one source of information. Now, whether that be for mental health, emotional needs, it could be, you know, more broadly when you think about current events, what they're seeing online, how to process information on social media. A majority is still saying we need support from parents. And they're not necessarily asking parents to just lecture them or talk to them, but listen more openly, be available to share thoughts and have an open dialogue.

Stephanie Hepburn 9:17

You have found that southern states lead in risk, and that includes Louisiana. I'm in New Orleans, Louisiana. Are you finding additional risk for youth who are turning to AI? Meaning the fact that they're not able to tap into resources as easily. Are you identifying youth in those states as more at risk for turning to AI for full support?

Shums Alikhan 9:38

We didn't specifically look at states when we looked at the pulse survey data around youth and AI. However, there is definitely a correlation. As I mentioned earlier, the youth that are using AI and they're turning to generative AI for mental health purposes, they overwhelmingly cited barriers to care. So transportation, cost, et cetera. And then there's a huge factor of provider shortages as well. So they don't know where to get the care because there are not that many mental health providers there. Our Thrive Atlas tool shows you where there is a higher risk for mental health provider shortages. And those southern states definitely jump out as some of the ones that have provider shortages, whether that's psychiatrists, psychologists, any mental health, or even school-based mental health specialists. They just don't have the mental health support. So they're turning to generative AI.

Stephanie Hepburn 10:33

In the show notes, I'll include a link to the Thrive Atlas. Is it powered by Generative AI?

Shums Alikhan 10:38

Yeah, so Thrive Atlas is a separate tool, essentially. It's part of the Youth Mental Health Tracker, but it is a powerful county level tool where we've brought in eight different external data sets and we've aggregated them down to the county level so that we can look at what is driving risk for youth. We look at limited wellness practice, provider shortages, accessibility barriers, socioeconomic hardship, negative life experiences, and limited support and belonging. These six risk indexes that factor into youth well-being and mental health. And this is aggregated down to the county level.

Stephanie Hepburn 11:17

So it's an index and analytics tool, is that right?

Shums Alikhan 11:20

Yes, that's correct. Does it utilize generative AI at all? We did use AI to a certain extent to aggregate that down and do the analytics to power the tool. A lot of the tools that are out there, they may provide a state level overview or they may provide just a general overview, but to take eight external data sets and aggregate it down to that level, you know, that's something that can essentially be done by AI.

Stephanie Hepburn 11:45

And much faster, I assume.

Shums Alikhan 11:47

Exactly. Shums, were you surprised by any of the findings? The biggest thing, and I guess it shouldn't be a surprise, but to me it was is that the 33% that are using generative AI as a substitute to getting care, it's because they're not getting professional help and they're using just generic tools in lieu of professional services. We also found that younger youth, boys, black and Hispanic youth, and youth experiencing financial hardships, they may be more likely to rely on generative AI as a substitute for professional care. But again, these youth are experiencing those barriers that I mentioned earlier as well. On the other hand, we did find that LGBTQ youth were a little bit more skeptic of generative AI in general. Now, this is talking more broadly, not just specific to mental health, but they were a little bit more skeptic and they were less likely to turn to generative AI. They cited concerns around safety, control, trust, and just general long-term impact. So we had sort of two groups. You know, we have this group that is engaging with AI for various reasons. And then we did find this proportion of LGBTQ plus youth that are less likely to actually turn to it. And they're citing concerns of safety, trust, just long-term overall impact to the environment as well.

Stephanie Hepburn 13:15

So you mentioned earlier that the large language models are fairly inconsistent as to whether they provide resources. I found that to be the case. And without standardization, that's going to continue to be the problem because these companies are determining internally what guardrails should be. What are your thoughts there, both on the state level, the federal, to ensure that people are provided resources?

Shums Alikhan 13:40

One of the things that came up as a potential solution, and of course, this would require a lot of discussions and a lot of time, but a potential simple safeguard would include a one-touch button next to the chat bar or attach the chat interface itself that will connect the user directly to human support services. So 988, text-based crisis services, NAMI. Having that up front and center, right next to the chat bar would be ideal. Secondly, what's really important is to co-create these resources. You know, include policymakers, include youth, include parents, include mental health experts to really get a deep dive and better understanding as to what would help and how youth are engaging with these tools. When we create these task forces, that's ultimately the best way to develop these safeguards and these guardrails. And youth want to do this. In previous surveys, they mentioned we want to be a part of the conversation. So it's our job to help them be a part of the conversation. And that's actually what we do as well is we co-create our surveys with the youth. We have a very iterative process where we go back and forth. So the same can be done when you're implementing safeguards and guardrails. Include them and evolve them as much as possible.

Stephanie Hepburn 15:05

So in developing the surveys related to AI, you have a youth task force that helps you develop those questions.

Shums Alikhan 15:13

We have a youth advisory board that we meet with almost on a monthly basis. And even before we create the survey, we actually talk to them about which topics are relevant, which topics matter to them the most. We really want to hear from them. And that's what makes the tracker more unique as well, is because this is data that you're getting from youth about youth that was co-created with youth. Meeting with them monthly allows that conversation to be ongoing. We start from the beginning in terms of what is the youth mental health ecosystem looking like right now? And then once we've developed the survey, we go back to them and say, okay, how do these questions sound? Do these resonate with you? And I think that's really important when you're creating any of these tools that are being used by youth.

Stephanie Hepburn 16:00

Were there specific concerns or questions that they've had about generative AI? Questions that they really felt were important to include in the survey?

Shums Alikhan 16:10

I think their main concern is reflected in our data. You know, generative AI is everywhere, it's developing rapidly. So, how can this be used correctly? What should we be doing? What does safe use look like for generative AI more broadly, too? I think the conversations have gone from looking at social media in isolation to now looking more broadly at the digital world, which includes generative AI.

Stephanie Hepburn 16:38

That's a great point. I mean, I think that's right that there's been a big shift from focusing on social media, though we are all still very much connected to social media. Do you find any differences when you're survey about social media versus generative AI in terms of what is an issue and how people are using it?

Shums Alikhan 16:58

So it's interesting. It's actually pretty similar. Um, mixed reviews, you know, it's a double-edged sword. It's providing some benefit in the sense of creating social connectedness, it's helping with isolation. However, there's also strong negatives as well. So very similar to generative AI in the sense that it is providing some benefit. We cannot say that all AI is bad. We cannot say all social media is bad because it's not. It's a double-edged sword. There are some positives, but then there are some negatives, and youth are seeing both sides as well.

Stephanie Hepburn 17:33

When you mentioned a button that should be available that connects users directly to mental health resources, is that something that should always be there? So if I go into, for example, ChatGPT or Claude, would it just be there hovering through every chat? Or are you saying that if implicit or explicit mental health language is part of the chat, then this button should pop up?

Shums Alikhan 17:57

I think it should always be there because sometimes the language may not be very explicit. You may start having a conversation with the language learning model and generative AI that sort of implies that you're in distress, but it may not be very explicit. You may have just had a bad day. You may have just broken up with your boyfriend or your girlfriend, and you're kind of starting to have these conversations, but generative AI may not pick up on that. So to have this one touch button always there is essential. And again, it goes back to the fact that a lot of these youth, they're saying that we don't even know where to get help. We don't know that these resources exist. So if that's up and centered, that hey, these are the resources that you have, this is what exists in terms of getting help, it would help tremendously. It should always be there.

Stephanie Hepburn 18:49

Also, the large language models themselves, ChatGPT, Claude, and Gemini are potentially a wealth of information on how many young people and adults are using large language models for mental health reasons. But right now, the searches are a black box. Is that something that you feel policymakers should be thinking about? It just seems like an untapped resource.

Shums Alikhan 19:17

Definitely. If that was something that was available, I think it would help tremendously. I think these surveys do a great job at capturing a good portion of what youth are saying and how they're using it. But of course, the more data we can get, the better. So if that was available, I think that would be great. Of course, that also brings into question, though, of privacy and trust issues. Youth did mention that generative AI is allowing them to sort of talk to an adult without having the concern of privacy and trust. So if we were to do something like that, I think there would have to be a disclaimer or there would have to be something put forth that, you know, this data is being used.

Stephanie Hepburn 19:59

But I assume the data is being used. You know, it's kind of like when you think about 911 call data, that is also a black box that we don't have enough information in terms of mental health calls, whether that's people calling 911 directly for themselves or for a third party. There's a lot of guesswork. What does this mean? How many calls? When we talk about 988, for example, one of the thoughts is well, we're guessing about how many calls could be diverted from 911. And there are locales where that information is transparent. And so they're their percentages. And then using those percentages to say maybe that's what it could be nationwide to divert 911 calls to 988, for example. When I think about the black box of these large language models, we are still on the guesswork side. We're guessing what these percentages are. And these surveys are extremely critical. I just think from a collaborative standpoint, that data could help drive your surveys. And it would be great for policymakers to know how many people are using these large language models for mental health purposes.

Shums Alikhan 21:12

Yeah, I think that's a great point. I think comparing it to 911 is also great, too. You're right. Youth are in putting all of this data, the data exists, seeing it in the forefront and allowing policymakers to see this, you know, would definitely put things into perspective, which is what we need to do.

Stephanie Hepburn 21:33

A lot of these companies, they say, we're not wellness companies, we're not mental health companies. That's not our aim and that's not our goal. And so what do you say to those companies? Because you're right, young people and adults are using large language models to help them navigate all kinds of issues. And that includes mental health, substance use, and other personal challenges that they may be navigating. Navigating. What do you say to them when they say, well, this is not our focus?

Shums Alikhan 22:06

I would show them the data. The data speaks volumes. This is what youth are saying. We have a third of our sample that turn to generative AI for mental health purposes as a substitute to even getting care. So I think that doing data collection like this is very important and it's critical because hopefully it can be eye-opening. Yes, it's not a mental health tool per se, but that's what many youth are using it for. So it may not have been intentionally created for that, but that's what it has become. And sometimes you have to evolve to that. You know, tech companies need to evolve to the changing landscape. For example, what Chat GPT was even a year or two years ago is not what it is right now. That requires changing your mindset.

Stephanie Hepburn 22:57

One thing that I thought was interesting when I was chatting with Gemini, it said it cared about my well-being. And so you have this shift that I've noticed. I don't know if you're noticing this, where there's an increase in anthropomorphizing AI. And I'm curious your thoughts when it comes to youth. Is that problematic? What needs to be done about that?

Shums Alikhan 23:24

Yeah, I think it's definitely problematic in the sense that youth are using this in lieu of getting other support services. So the more we humanize these tools, the more it's going to seem like they're talking to a human and they're going to defer away from getting additional professional support. There's a fine line, and I can see the benefits, but I can also see the risks.

Stephanie Hepburn 23:50

So you don't want a young person who is using one of the large language models to help them figure out how to, for example, talk about their mental health issues with their parents, where they're getting some feedback. You don't want that young person to develop a parasocial relationship with AI in concern that not only is that bonding a human with AI, which is not human, what you're saying is it can actually make it more challenging for the person to want to turn away from the AI and to human connection.

Shums Alikhan 24:28

Exactly. It'll make it more challenging for them to go to the human connection, which is what we ultimately want. We still want them to find human support, whether that's through school support systems, getting professional mental help, uh, care through counselors or psychiatrists, psychologists, whatever that may be, professional support is still ultimately going to trump these tools when it comes to getting mental health support specifically. So we definitely don't want to deter that.

Stephanie Hepburn 24:57

Shums, what's next? Do you plan on doing another survey regarding young people and mental health? What's the next step?

Shums Alikhan 25:06

We're doing another national big round of data collection to compare results from our 2023, 2024 first national survey. And what we'd like to do is really compare the data and see where are the trends, what has changed, what has improved, what has gotten worse in terms of overall youth mental health and well-being. You know, where has the conversation shifted? And I'm sure generative AI will come up in there because a couple of years ago, this was not even a conversation many people were having. So that's definitely something that we will take into consideration and incorporate into our big national survey. And that will be up and released later in the fall.

Stephanie Hepburn 25:48

Thank you so much for coming on, Shums.

Shums Alikhan 25:51

Thank you for having me.

Stephanie Hepburn 26:09

During this episode, I liken the lack of transparency of AI companies regarding mental health data to the black box of 911 call data, specifically calls for help that could have been diverted to 988, the nationwide number for mental health and substance use crisis. As with 911 call data, I highlight that AI companies are sitting on large-scale strip-down data that could help us better understand the scope of how people are using large language models to help address their mental health concerns. In October, OpenEI released that around 0.07% of Chat GPT users active in a given week exhibit possible signs of mental health emergencies. That includes mania, psychosis, or suicidal thoughts. I'm your host and producer. Our associate producer is Ren Koenig, Audio Engineering by Chris Meehan, Music is Vinyl Couch by Blue Dot Sessions. If you enjoyed this episode, please subscribe and leave us a review wherever you listen to the podcast. It helps others find the show. Thanks for listening.