The Future of Mental Healthcare: What AI Can and Can't Do

“I’d be comfortable using AI as long as a real doctor is reviewing it.” That’s how many people feel when they think about AI and their mental health, and the research has landed in a similar place. The need behind the hype is real. According to NIMH survey data, 59.3 million U.S. adults had a mental illness in 2022, and only 50.6% received any treatment that year. AI’s appeal is precision at scale: it can ask structured questions the same way every time, at any hour, for anyone who opens the app.

The current standard of care works when people can get it. A 2018 network meta-analysis of 522 trials found that all 21 antidepressants studied were more effective than placebo for adults with major depression. The results are real, and the science is well established.

But that standard still reaches only about half the people who need it, and at Siggy we think AI’s job is to extend licensed care rather than imitate it.

This guide covers why AI is moving into mental health, what the trials show, where it falls short, how regulators are responding, and how to judge a tool before you trust it. Clinical oversight is what separates helpful AI from risky AI in mental health, and that conclusion comes straight from the evidence.

What This Page Covers

• Why Is AI Moving Into Mental Health Care?
• What Does the Evidence Say About AI Chatbots?
• Where Does AI Fall Short?
• How Are Regulators Responding?
• Why Does Engagement Matter So Much?
• How Can You Judge an AI Mental Health Tool?
• Start With AI That Answers to a Clinician

Why Is AI Moving Into Mental Health Care?

People are already using it. The American Psychological Association’s 2025 health advisory notes that emotional support is one of the most common uses of generative AI chatbots, even though those chatbots weren’t created to deliver mental health care. The advisory links that trend to the wider mental health crisis, rising loneliness, and too few providers to meet demand.

The appeal makes sense. AI is available at 2 a.m., costs little, and doesn’t make you wait months for a first appointment. Some people also find it easier to open up. “I think I would feel less embarrassed talking to AI about it” is a common sentiment.

The open question is how to get the benefits without the harms. Our post on AI and mental health covers the basics of how these tools work.

What Does the Evidence Say About AI Chatbots?

The evidence is promising but early. A 2024 meta-analysis of 18 randomized trials with 3,477 participants found that AI chatbots produced small improvements in depression (effect size 0.26) and anxiety (0.19). Benefits were clearest after about 8 weeks, but at a 3-month follow-up, the authors found no substantial effect for either condition.

Generative AI, the technology behind tools like ChatGPT, has less evidence. The first randomized trial of a generative AI therapy chatbot, Dartmouth’s Therabot, enrolled 210 adults with clinically significant depression, anxiety, or eating disorder risk. People who used the chatbot for four weeks had larger symptom reductions than a waitlist group, and they rated their bond with it as comparable to the bond with a human therapist. The authors said larger studies are needed, and Dartmouth’s research summary noted that AI-powered therapy still needs clinician oversight.

That’s real progress, with an important caveat. The comparison group got no treatment at all. An FDA advisory committee later noted that waitlist controls may not be ideal for studies of these devices, and the APA advisory says future studies must move beyond them. We don’t yet know how a chatbot compares with a human clinician.

Where Does AI Fall Short?

The risks show up most in the moments that matter most. A Stanford-led study tested large language models against standards drawn from clinical therapy guides. The models expressed stigma toward people with some mental health conditions and responded inappropriately to some common and critical situations, including encouraging delusional thinking. Newer and larger models showed the same problems, and the authors concluded these models should not replace therapists.

The APA advisory adds that using chatbots and wellness apps for mental health can have unintended effects and even cause harm. It describes these tools’ ability to consistently and safely manage someone in crisis as “limited and unpredictable,” and it notes that even purpose-built AI tools still lack high-quality, large-scale clinical trials showing they’re effective and safe. It also flags sycophancy, the tendency of many chatbots to agree with and validate whatever a user says, as a particular risk in mental health.

Our comparison of AI therapy vs. human therapy goes deeper on what a human clinician brings that software can’t.

How Are Regulators Responding?

Regulators are moving toward a clear principle: AI can extend care, but people need to stay in the loop. In November 2025, the FDA’s Digital Health Advisory Committee met to discuss generative AI mental health devices. Here’s how its 24-hour meeting summary frames the balance.

Area	Where AI Can Help	What Still Needs a Clinician
Access	Earlier, broader access, including rural areas	Full evaluation before the device is used
Monitoring	Tracking response and triaging symptoms	Review of the data by the prescriber
Crisis	Single-tap access to more support	A plan with your clinician for escalation
Severity	More comfort with mild symptoms	Moderate to severe depression

The committee also agreed that users should be told clearly that the AI system isn’t a human therapist, and that labeling should include crisis information, overuse warnings, and data and privacy details. Members flagged risks including hallucinations, missed therapeutic cues, worsening symptoms such as self-harm, and unhealthy attachment to the chatbot.

Why Does Engagement Matter So Much?

A tool only helps if people keep using it, and mental health apps have struggled with that. In an analysis of 93 apps for anxiety, depression, and emotional well-being, the median 30-day retention rate was just 3.3%.

The chatbot meta-analysis tells a similar story from another angle: benefits faded by three months. The FDA committee noted that the relationship between engagement, treatment adherence, and outcomes isn’t well understood yet. More time in an app isn’t automatically better care.

What seems to matter is continuity: someone who knows your history, notices when things change, and adjusts your care. That’s the gap AI could help fill, but only when it’s connected to clinicians who can act on what it sees. Our post on clinician-supervised apps vs. unsupervised AI breaks down that difference.

How Can You Judge an AI Mental Health Tool?

Use the safeguards regulators and researchers keep pointing to. Before you trust a tool with your mental health, ask:

• Does a licensed clinician evaluate you and review treatment decisions?
• Does the tool clearly tell you that you’re talking with AI rather than a person?
• Can you reach a human and crisis resources quickly?
• Has it been tested in randomized trials with a real comparison group?
• Does it explain in plain language how your data is used?
• Is it honest about what it handles, and what it doesn’t?
• Will you tell your clinician which AI tools you use? The APA strongly recommends it, so guidance that conflicts with your treatment plan gets caught.

The APA advisory also says chatbots and wellness apps shouldn’t replace a qualified mental health provider, though they may be appropriate as a supportive add-on to ongoing care. If you’re in crisis, call or text 988. In an emergency, call 911.

This article is for education only and isn’t a substitute for personal medical advice. It reflects sources available as of October 2026.

Start With AI That Answers to a Clinician

The future of mental health care is most likely AI working alongside licensed clinicians, making care easier to reach and harder to fall out of. Siggy is built around that idea. Start your anonymous intake with Siggy to talk through what’s going on in your own words, at your own pace. It’s free, asks for no name or email, and gives you a downloadable summary you can share with any clinician. Siggy combines human clinicians with AI therapy and daily check-ins, and its licensed prescribers diagnose and treat the full range of mental health conditions. A licensed prescriber reviews your full clinical picture before anything is prescribed. To see how that review works, read our guide to what clinically supervised care means.