How to spot red flags in mental health apps

Not long ago, patients might have told their therapists about a self-help book that inspired them or a conversation with a friend that unnerved them. And while these disclosures still occur, they’re increasingly accompanied—if not replaced—by mentions of the AI tools or apps leaned on between sessions.

Patients are “very much using ChatGPT to bounce things off of and see, What does my therapist say? What does Chat GPT say?” says Rachel Wood, PhDopens in new window, a cyberpsychology researcher and therapist in Colorado Springs, where she leads the AI Mental Health Collectiveopens in new window to connect clinicians over how AI is affecting their practices and patients’ lives.

It’s not just ChatGPT. According to one 2024 survey in BMC Public Healthopens in new window, 41% of people in the general population said they’d used a mental health app, like for meditation or mood-tracking, in the past 12 months. Other mental health apps, like those developed for a particular condition or to supplement therapy, are increasingly entering the market—and clinicians’ offices—too.

Many psychologists have mixed feelings. On the one hand, such apps can help patients more seamlessly integrate therapeutic principles in their lives—prompting those with anxiety to do exposure therapy homework, for instance, or guiding those with insomnia toward healthy sleep habitsopens in new window in real time. One 2024 randomized clinical trialopens in new window also found that apps rooted in cognitive behavioral therapy can improve anxiety and depression symptoms before people start therapy, acting as a worthwhile bridge while waiting for care.

“As a psychologist, the usual mode of delivering care is maybe seeing 20, 25 patients a week,” said Eric Kuhn, PhDopens in new window, an associate professor of psychiatry and behavioral sciences at Stanford Medicine who’s been developing mobile mental health apps with the U.S. Department of Veterans Affairs for about 15 years. “So, to have an impact where you have 5 million downloads, it’s super exciting to see—and gratifying.”

On the other hand, many if not most apps raise privacy concerns, lack rigorous testing, and aren’t uniformly regulated or evaluated. In some cases, what’s a helpful tool for one person can border on harmful for another. “As a researcher, I think there’s a lot of potential benefit,” said Margaret Morris, PhDopens in new window, an affiliate associate professor at the University of Washington, where she studies how technologies including AI can help patients and therapists. But as a clinical psychologist in private practice, she added, she’s more cautious.

Still, it’s important for psychologists to understand the landscape because patients aren’t waiting for psychologists’ approval to experiment. “Hundreds of millions of people are [using these apps],” Wood said. “So, my position is: How do we steer and guide the safest experience?” Here’s what she and other experts think works best.

Before evaluating the tools themselves, psychologists recommend getting a picture of patients’ digital lives. That could mean adding a question to intake assessments like, “What role, if any, do AI, chatbots, or mental health apps play in your life?”

“We ask about every other support area of their lives, like, What role do hobbies play? What about family and friends and faith communities?” Wood said. “[Digital tools] are becoming a major part of people’s ecosystems, and so we need to ask a question like this in a nonjudgmental way.”

Thekla Brumder Ross, PsyDopens in new window, a clinical psychologist in Milwaukee with expertise in digital harm reduction, recommends asking patients about how much of their lives are “URL versus IRL.” Therapists can even look, with their patients, at information like the phone’s total screen time, the most used apps and the number of phone pickups. That data, alongside information about the person’s offline social life and hobbies, can reveal a “full digital landscape,” Brumder Ross said.

“Understanding that balance [between online and offline life] frames whether app-based interventions will support or potentially undermine their mental health goals in their treatment plans,” she added.

For patients who are already using mental health apps or AI-based emotional support, therapists should guide the conversation with curiosity, Morris suggested. When and why do they most often use various tools? What are they getting from it they don’t get elsewhere, including in therapy? What have they learned, and who else do they tell about what they’re discovering?

“My stance as a therapist is really listening to what people are already using, and then trying to learn from that what they’re looking for help on,” Morris said. “I want to understand if they are using AI in a way that brings them closer to others or worsens isolation.” For example, maybe they’re asking an AI chatbot about physical health frustrations or using a tool, consciously or not, to avoid real-life confrontations and connections. Maybe they’re using a wellness app that’s actually reinforcing disordered behavior.

For example, someone could be “getting really obsessed about physical activity or sleep, when the person’s real challenges are more about cognitive flexibility,” Morris said. “That’s another case where someone might be using these tools in a way that feeds into a problem, rather than working on their true issues.”

In other words, no matter how well-designed an app is or how strongly it’s backed by psychological research, it can still raise red flags when used in the wrong situation. It’s up to clinicians to keep those conversations going.

“Almost any tool can be used in a way that’s harmful, and a lot of tools that we don’t think about as mental health tools can be used in ways that are helpful,” said Morris, whose book, Left to Our Own Devices, explores how people can adapt technology to support their relationships and health. “We have agency in how to use those tools, and talking about it with patients helps activate that agency.”

Of course, some mental health apps are simply better, safer, and more effective than others. But keeping them all straight can be extremely difficult. One paper states that there were more than 10,000 on the marketopens in new window in early 2023 and another found 20,000opens in new window. “With such a proliferation of apps, [vetting them] is a task that is impossible to keep up with,” said Kuhn, who’s a member of APA’s Mental Health Technology Advisory Committee.

The products are wide-ranging in their purposes and the oversight around them, too. For instance, many apps that are considered “general wellness,” like those that offer guided meditations or help people track habits, are not regulated by the U.S. Food and Drug Administration (FDA). Others, like the depression app Rejoynopens in new window and the anxiety app DaylightRXopens in new window, are considered digital therapeutics and are regulated by the FDA because they deliver a treatment for a medical condition. As part of the regulatory process, the device makers submit data from RCTS to support the safety and effectiveness of the tool.

That said, just because an app isn’t FDA-cleared doesn’t mean it’s no good. Many simply don’t meet the risk threshold for the organization to assess them, according to Stephen Schueller, PhDopens in new window, a professor of psychology and informatics at the University of California, Irvine, who studies how to leverage technology to improve access to mental health services. “For clinicians and people who are interested in using these, it’s not easy to parse this landscape,” Schueller said.

But some emerging tools and frameworks can help. Last year, APA rolled out a step-by-step guide to help practitioners decide which AI tools are right for their practice. The American Psychiatric Association’s Evaluation Modelopens in new window, also walks users through a rating system to help them make a thoughtful choice about app selection. And most recently, APA launched its Digital Badge Program to signal the digital mental and behavioral health tools that meet certain criteria in clinical value, regulatory compliance, user safety, and data privacy.

“Over time, we’ve seen different organizations try to vet the apps and give consumers, clinicians, and the public some ideas about which apps are good, and which aren’t,” Kuhn said. In total, these guides point to “a few rules of thumb clinicians can use to make sure that the apps that they’re recommending are going to be helpful and not harmful. That’s where the field largely landed.”

Short of, or in addition to, an official evaluation system, experts say therapists can watch for these red flags when coming across a mental health app in their clinics—or in the digital wild.

A lack of privacy protections

In one of Schueller’s studiesopens in new window, published in Internet Interventions, he and colleagues evaluated 116 mobile apps for depression. When reviewing their data security and privacy policies, they found only five apps received a transparency score of “acceptable,” while the majority (79 apps or 68%) were “unacceptable.” Just over half had no privacy policy at all. “So, they didn’t tell you anything about what they do with the data,” Schueller said.

That’s why Wood recommends looking toward apps with a clear purpose, like helping treat insomnia, rather than general wellness tools that more often lack privacy protections like HIPAA compliance. “What’s the true cost of using a tool?” she asked. “It might be free to start, but what are you not getting—and what are you giving away?”

No crisis intervention

Any app someone is using for mental health support should have a clear way to connect to real-world help, Wood said. “There’s got to be a direct way to get to a human helpline very quickly, where you’re off the app and you’re talking to someone because you may be in crisis,” she said.

An inability to download your own data

Apps should also give users ownership over their own information. “If someone is using an app for tracking something like their sleep or fitness, I want them to be able to make that data their own and look at it in conjunction with other things,” said Morris, who, along with colleagues at the University of Washington and Columbia University, has been studying how AI can help clinicians analyze data from watches and phones to understand patient’s sleep, activity, and other patterns. “If they want to be able to share it within the context of therapy, something that’s totally insular and locked down seems less useful.”

It may also be impractical if the app is intended to be a part of the patient’s care plan, Brumder Ross said. “If you’re recommending an evidence-based, clinically safe, data-secure app, that’s an extension of the trust in your relationship,” she said. “So you want be sure that that data is going be used and clinically informed in the treatment.”

Overblown claims

A reputable app should also be upfront about what it can and can’t provide. And if the product does say it’s rooted in a certain psychological principle or therapeutic technique, that method should be front and center, and accurate.

“You have to be careful because they can say, ‘Cognitive behavioral therapy works and it works in digital formats,’ and that’s great,” Schueller said. “But if the app doesn’t actually deliver cognitive behavioral therapy, it’s not so beneficial.”

A spotty evidence base

Likewise, while most mental health apps claim they’re scientifically sound, few actually mean the product itself has been proven to do what it says it does. In fact, one of Schueller’s studies found that less than 4% of anxiety apps had undergone a randomized clinical trialopens in new window for feasibility and efficacy.

“That’s not to say there’s not a strong evidence base that these things work—there are hundreds of randomized controlled studies that demonstrate that digital mental health can be effective and impactful,” Schueller said. But for consumers, there’s an easy-to-overlook difference between evidence-based and evidence-informed, Kuhn agreed. “A lot of times you’ll see apps that say, ‘this is evidence based,’ and then you go to the part of their website that shows the evidence base and it has nothing to do with the app,” he said.

No mental health expert or organization involved

Apps backed by clinicians can add a vote of confidence, Brumder Ross finds. While psychologists’ goals are likely improved mental health and well-being outcomes, product developers’ goals are often to keep users engaged. “If the app’s measure of success is time on platform, the entire motivation is keeping your attention for as long as possible,” she said. “So, I like to look at apps that have clinicians who have informed their treatment and their protocol.”

Kuhn recommends looking into whether the app was developed by or in partnership with a reputable organization like a university, hospital, or the VA.

Poor functionality

Kuhn prioritizes apps that are available on both iOS and Android, and that don’t need internet access after being downloaded. “This is good for folks who might be in rural areas or for military service members who are deployed,” he said.

It’s helpful, too, to know how recently the app was last updated; if it’s been more than six months or a year, it could have been abandoned and at higher risk of security breaches. Ease of use matters too. “Is it pleasant to use? Can it be accessed for users who have disabilities? What’s the reading level? How tech savvy does the user need to be to be able to use the app?” Kuhn said to ask. “If it’s difficult to navigate, if the user interface isn’t pleasant, isn’t engaging, folks will just stop using it.”

An app that acts more like a friend than a therapeutic tool

Apps—particularly those that integrate AI-enabled chats—should stick to their purpose (say, helping patients tackle twice-daily CBT exercises) and stay out of “companion” territory. Bots that float between friend, colleague, family member, and even love interest can be “incredibly disorienting” to people seeking mental health support, Wood said.

“Could you imagine if your therapist just started flirting with you? This is ethically very, very wrong,” she said. “It’s the same thing for these apps—you want to make sure it doesn’t become fluid in its role, but it’s simply there to say, ‘I am here to help you with your grounding exercises or your reflective journaling, and that’s it.’”

Good fit-for-purpose models, by contrast, “give clear reminders of nonpersonhood, meaning they’re letting you know, hopefully at regular intervals, ‘Hey, I’m an app, I’m a machine,’ or ‘I’m a large language model. I’m not a person,’” Wood said. “They should not be simulating relationships as much as they should simply be helping the development of certain skills.”

A quality mental health app shouldn’t sweet-talk users into lingering longer either. But some research shows 37% of farewells analyzed across relational apps use manipulative tacticsopens in new window like enticing them into engaging more (“Going so soon? One more thing!”), even after the user has indicated their intent to leave. “That’s another huge red flag,” Wood said. Like a therapist seeing a patient out at a session’s close, an ethical app simply says, “Bye.”