Covered by insurance in all 50 states, incl. Medicare Advantage day one
Siggy
Back to Blog

Clinically Supervised AI in Mental Health: What the Clinician Actually Does

Clinically reviewed by Wendy Delgado, PA-C

Reviewed November 17, 2026 · Last updated November 17, 2026

Key Takeaways

  • Clinically supervised AI should mean a licensed clinician reviews what the AI produces and makes every clinical decision, including diagnosis, prescribing, and treatment changes.
  • General-purpose chatbots are least reliable where stakes are highest. In a 2025 study of 9,000 responses, three popular chatbots didn't meaningfully distinguish among intermediate levels of suicide risk.
  • In the Therabot trial, clinicians reviewed every chatbot response and had to step in 13 times, including to correct inappropriate medical advice.
  • Rules are moving toward human accountability. Illinois restricts AI use in clinical mental health care, and FDA advisors have stressed safety monitoring and human escalation paths for crises.
  • Before you trust an app, ask who makes the diagnosis, who approves prescriptions, and what happens in a crisis. If you're in danger, call 911, or call or text 988.

Maybe you’ve typed something into a chatbot at 2 a.m. that you’ve never said out loud to anyone. Maybe it helped. And maybe a quieter question followed: is anyone actually checking what this thing tells me? Plenty of people land in the same place. They’d be comfortable using AI for their mental health, as long as a real doctor is reviewing it.

That instinct matches how clinical judgment works. Mental health care runs on accountability: a licensed person gathers your history, weighs what it means, and answers for the decision that follows. Getting that right takes precision, because the same sentence, like “I can’t keep doing this,” can signal exhaustion in one person and danger in another. Clinical supervision is the structure that puts a trained human at that decision point.

The traditional model places that judgment with a clinician, supported by validated tools. The PHQ-9 is a nine-question depression measure used across primary care and psychiatry. In its validation study, 6,000 patients completed it, and in the 580 who were also interviewed by a mental health professional, a score of 10 or higher identified major depression with 88% sensitivity and 88% specificity. The results are real, and the science is well established.

The hard part is reaching that clinician. In a 2023 mystery shopper study of 948 psychiatrists in five states, only 18.5% were available to see new patients, and the median wait for an in-person visit was 67 days. AI can help close that gap, and at Siggy we only use it in a way that keeps a licensed clinician accountable for every clinical decision.

Clinically supervised AI, done properly, splits the work by risk. Software handles the structured parts, like collecting your history and organizing screening scores, while a licensed clinician reviews that material, makes the call, and stays accountable for it. It solves the same problem the traditional model solved, an accurate assessment, with less waiting and the same human sign-off. The difference comes down to who answers for the decision, and that’s something you can check.

What This Page Covers

What Does Clinically Supervised AI Actually Mean?

Clinically supervised AI means a licensed clinician reviews what the AI produces and makes every clinical decision, including diagnosis, prescribing, and changes to treatment. The AI can gather and organize information, but it doesn’t make the call.

Used that way, AI is good at the parts of care that are structured and time-consuming. It can walk you through a detailed intake at whatever hour you’re ready, ask every standard screening question without rushing, and organize your answers so a clinician can read your story in minutes instead of rebuilding it from scratch.

The phrase gets used loosely. Some apps use it because a clinician helped write the chatbot’s scripts. Others mean a clinician is available if you ask for one. Neither arrangement puts a licensed person between the AI and a decision about your care. A useful test is to ask: if the AI got something wrong about you today, which human would see it, and when? Our side-by-side look at a doctor-supervised app vs. unsupervised AI covers how those models differ day to day.

Why Does Human Oversight Matter So Much in Mental Health?

General-purpose chatbots are least consistent where the stakes are highest. In a 2025 Psychiatric Services study, 13 clinical experts rated 30 suicide-related questions by risk level, and researchers asked ChatGPT, Claude, and Gemini each question 100 times, for 9,000 responses in all. ChatGPT and Claude lined up with the experts at the extremes, always answering very low-risk questions directly and never answering very high-risk ones directly, while Gemini was more variable. Between those poles, the chatbots didn’t meaningfully distinguish low, medium, and high risk.

The problems go beyond suicide risk. A 2025 ACM conference paper tested large language models in therapy-style conversations and found they expressed stigma toward people with mental health conditions. They also responded inappropriately in some crisis scenarios and sometimes encouraged delusional thinking, a failure researchers tie to the models’ tendency to agree with the user. Newer and larger models didn’t clearly do better, which suggests scaling up alone won’t fix it.

None of this rules AI out of mental health care, but it does mean a trained human should check a chatbot’s output before it shapes a decision about you.

What Does the Research Say About AI With Clinician Oversight?

The early evidence is encouraging, with real limits. A 2023 meta-analysis of 15 randomized trials found that AI conversational agents reduced symptoms of depression and distress with moderate effect sizes, but they didn’t significantly improve overall psychological well-being.

The first randomized trial of a fully generative AI chatbot for clinical-level symptoms, according to its authors, was a 2025 NEJM AI study of Therabot, built at Dartmouth. It randomized 210 adults with significant symptoms of depression, anxiety, or eating-disorder risk to four weeks of use or a waitlist. People with depression symptoms improved significantly more than the waitlist group at four and eight weeks. The comparison was a waitlist rather than a human clinician, so the trial shows the chatbot outperformed waiting, and it doesn’t show the chatbot matched a clinician.

How the trial was run matters as much as the result. According to a 2026 scoping review of 21 studies, clinicians supervised every Therabot response and had to step in 13 times, including to correct inappropriate medical advice. The trial also excluded people with suicidal thoughts, severe depression, mania, or psychosis. Across the 21 studies, ongoing human oversight was uncommon, crisis protocols were mostly underdeveloped, and documented failures included missed suicidal ideation. The review’s authors also note that reviewing every message by hand is unlikely to scale, which is why the real design question is where human judgment sits, and who owns each decision.

What Do Regulators Expect From AI Mental Health Tools?

The rules are moving toward human accountability. As a JAMA Health Forum commentary notes, Illinois’ Wellness and Oversight for Psychological Resources Act restricts AI use in clinical settings. A 2026 policy review describes the law as barring fully autonomous AI from delivering mental health treatment and keeping that role with licensed professionals.

Federal regulators are paying attention too. The FDA’s Digital Health Advisory Committee has examined generative AI mental health devices, and the 2026 scoping review notes that its recommendations stressed post-market safety monitoring, clear definitions of AI-specific adverse events, and human escalation pathways for crises. A 2026 review in Frontiers adds that the committee’s 2025 materials counted more than 1,200 AI-enabled medical devices authorized by the FDA overall, with none authorized for mental health uses.

What Does the Clinician Actually Do in a Supervised Workflow?

In a well-built workflow, the clinician’s role shows up at four points.

  1. Before any decision, the clinician reviews the full intake, including the conversation itself and validated scores like the PHQ-9, rather than relying on a one-line summary.
  2. At the decision, a licensed clinician makes and documents any diagnosis or prescription. The AI’s suggestions are input the clinician can accept, change, or reject.
  3. Between visits, the AI may collect check-ins or side-effect reports, but anything clinical goes to a human to act on.
  4. In a crisis, the person is pointed to emergency help, 911 or 988, because software shouldn’t be anyone’s only safety net.

If any of those four points has no named human attached, the supervision is thinner than the label suggests. Our guide to what psychiatrist oversight of AI chat should look like goes deeper on that standard.

How Can You Tell if an AI Mental Health Service Is Really Supervised?

These questions turn the research above into a quick screen, and a trustworthy service should answer each one plainly.

Question to Ask What a Strong Answer Sounds Like
Who makes the diagnosis? A licensed clinician
Who approves prescriptions? A licensed prescriber, every time
Does a human see the full conversation? Yes, more than a summary
What happens in a crisis? Clear direction to 911 or 988
Does the app say it’s AI? Clearly, from the start
What happens to my data? Plain-language privacy terms

Vague answers, or answers that only mention a clinician “available on request,” are worth pressing on before you share anything sensitive.

How Siggy Fits In

Siggy is built around that split. Siggy’s AI runs your intake conversation, covering symptoms, history, current medications, and validated screening questions like the PHQ-9. A licensed prescriber reviews your full intake, including the conversation itself, before anything is prescribed. The AI doesn’t diagnose you, and it doesn’t choose, start, stop, or change your medication. Clinicians treat what it prepares the way they’d treat a chart prepared by staff: as input they can accept, change, or override.

Between visits, Siggy offers coping exercises rooted in clinical research, ready when a hard moment hits, like paced breathing or grounding. Siggy’s AI isn’t a therapist and doesn’t provide therapy, so if you already see a therapist, that relationship stays yours. The intake asks for no name, email, or account, so you can be honest without worrying how it sounds, and you can download your summary to share with anyone, including your therapist or primary care doctor.

Siggy isn’t a crisis service. If you’re in danger, call 911, or call or text 988. This article is for education only and isn’t a substitute for personal medical advice. It reflects sources available as of November 2026.

Start Your Anonymous Intake

Clinically supervised should mean one thing: a licensed clinician sees what the AI produced and owns every decision about your care. The research and the regulators point the same way, toward AI that prepares and humans who decide. For a closer look at how that standard plays out in online care, see our guide to what “clinically supervised” means in online psychiatric care.

If you’re in crisis right now, call 911 or call or text 988 first. If your mental health has been weighing on you and you want care where a real person signs off on every decision, start your anonymous intake. It’s free, there’s no name or email required, and a licensed prescriber reviews your intake before anything is prescribed. Your summary is yours to keep and share.

Mental healthcare should stay with you between appointments.

Siggy combines daily check-ins with clinician-supervised care so your treatment plan can respond to what is actually happening.

Start anonymously. A real doctor reviews every clinical decision. HIPAA-compliant.

Start Anonymous Intake

Frequently Asked Questions

What does clinically supervised AI mean in mental health?

Clinically supervised AI means a licensed clinician reviews what the AI produces and makes every clinical decision, including diagnosis, prescriptions, and treatment changes. The AI can gather and organize information, such as your history and screening scores, but a human clinician decides what happens next and is accountable for it.

Is it safe to use an AI chatbot for mental health without a clinician involved?

The evidence doesn't support relying on AI alone. A 2025 study found three popular chatbots were inconsistent with intermediate-risk suicide questions, and a 2026 review of 21 studies of mental health chatbots found that ongoing human oversight was limited and crisis protocols were mostly underdeveloped. That's why a licensed clinician in the loop matters.

What does Illinois law say about AI in mental health care?

Illinois restricts it. The state's Wellness and Oversight for Psychological Resources Act limits how AI can be used in clinical mental health settings and bars fully autonomous AI from delivering mental health treatment, keeping that role with licensed professionals.

Can an AI chatbot diagnose depression or prescribe medication?

In a properly supervised model, no. AI can collect symptoms and score validated questionnaires like the PHQ-9, but a licensed clinician makes the diagnosis and any prescribing decision. If an app suggests its AI diagnoses or prescribes on its own, treat that as a red flag.

How can I tell if a mental health app is really supervised?

Ask who makes the diagnosis, who approves prescriptions, whether a human reviews the full conversation or only a summary, what happens in a crisis, and whether the app clearly says it's AI. Strong answers name licensed clinicians and point you to 911 or 988 in an emergency.

Is Siggy a replacement for a therapist?

No, Siggy's AI isn't a therapist and doesn't provide therapy. It runs a structured, anonymous intake, and a licensed prescriber reviews your full intake before anything is prescribed. Between visits, Siggy offers research-grounded ways to cope, and you can share your downloadable intake summary with a therapist you already see. Siggy isn't a crisis service, so call 911 or call or text 988 in an emergency.

Sources

  1. Kroenke K, Spitzer RL, Williams JB. The PHQ-9: validity of a brief depression severity measure. Journal of General Internal Medicine. 2001;16(9):606-613. doi:10.1046/j.1525-1497.2001.016009606.x
  2. Sun CF, Correll CU, Trestman RL, et al. Low availability, long wait times, and high geographic disparity of psychiatric outpatient care in the US. General Hospital Psychiatry. 2023;84:12-17. doi:10.1016/j.genhosppsych.2023.05.012
  3. McBain RK, Cantor J, Zhang LA, et al. Evaluation of alignment between large language models and expert clinicians in suicide risk assessment. Psychiatric Services. 2025;76(11):944-950. doi:10.1176/appi.ps.20250086
  4. Moore J, Grabb DJ, Agnew WS, et al. Expressing stigma and inappropriate responses prevents LLMs from safely replacing mental health providers. ACM Conference on Fairness, Accountability, and Transparency (FAccT ’25). 2025. doi:10.1145/3715275.3732039
  5. Li H, Zhang R, Lee YC, Kraut RE, Mohr DC. Systematic review and meta-analysis of AI-based conversational agents for promoting mental health and well-being. npj Digital Medicine. 2023;6(1):236. doi:10.1038/s41746-023-00979-5
  6. Heinz MV, Mackin DM, Trudeau BM, et al. Randomized trial of a generative AI chatbot for mental health treatment. NEJM AI. 2025;2(4):AIoa2400802. doi:10.1056/AIoa2400802
  7. Olisaeloka L, Richardson CG, Wang AY, Munthali RJ, Vigo D. Safety mechanisms and risk mitigation in generative AI mental health chatbots: a systematic scoping review. Healthcare. 2026;14(10):1395. doi:10.3390/healthcare14101395
  8. Shim RS. Why AI’s mental health risks demand regulation now. JAMA Health Forum. 2026;7(7):e263056. doi:10.1001/jamahealthforum.2026.3056
  9. Schoene AM, Mestre R, Middleton SE, et al. Responsible AI in mental healthcare: policy directions and stakeholder insights. Frontiers in Public Health. 2026;14. doi:10.3389/fpubh.2026.1814039
  10. Srivastava K, Srivastava R. Explainable artificial intelligence (XAI) in schizophrenia: enhancing transparency in clinical decision-making. Frontiers in Systems Biology. 2026;6. doi:10.3389/fsysb.2026.1898853