In short
The best use of AI in mental health is around therapy, never instead of it. Before you start, AI can help you name what you are dealing with, find words for your experience, and prepare questions for a clinician. Between sessions, it can support skill practice, mood logging, and homework. Afterward, bring what you noticed back to your therapist and keep them in the loop. Crisis moments, medication decisions, and diagnosis never go to an app. If you are in crisis or thinking about suicide, call or text 988 (US Suicide & Crisis Lifeline), available 24/7.
The short answer: AI works best around therapy
If you are already in therapy or headed there, the most useful way to think about AI is as scaffolding around the work. It can help you arrive prepared, practice between appointments, and reflect afterward. The therapy itself, the relationship, the clinical judgment, and the accountability, stays with your licensed therapist.
That framing matters because the evidence points hard at the relationship. The largest meta-analysis on the topic, covering 295 studies and over 30,000 patients, found that the alliance between client and therapist is a robust predictor of how well therapy goes, across different treatment types and settings. A chatbot can hand you a worksheet at midnight. It cannot form that alliance, and it cannot take responsibility for your care. Understanding the difference between an AI vs human therapist helps you place each tool where it belongs.
This is also where the field is heading in practice. In an American Psychological Association survey of 1,242 licensed psychologists conducted between April 9 and April 26, 2026, 77 percent said they had spoken with patients who use AI for support, engagement, or other reasons. People are already using these tools around their therapy. The playbook below is about doing it deliberately.
My rule after months of testing these tools is simple: the app gets the prep and the practice, the therapist gets the decisions. The moment I caught a chatbot happily giving me the same reassurance a fourth time in a row, I understood the trap. Log the pattern, bring it to a human, and the tool goes back to being useful.
Before therapy: get to the right door, prepared
The stretch before your first session, or before you have even chosen a therapist, is where AI is at its most genuinely useful. Many people delay care for months because they cannot name what they are dealing with or do not know what kind of help to ask for.
Use a chatbot to find words for your experience. Describe what your days feel like and ask it to reflect the themes back to you. Say you keep replaying conversations at 2 a.m., or that you have stopped answering texts. Seeing that summarized in plain language can turn a vague sense that something is wrong into something you can say out loud to a professional.
Use it to learn the landscape. Ask what the difference is between a psychologist, a counselor, and a psychiatrist, what CBT involves, or what a first session usually looks like. Ask what questions are reasonable to bring to a consultation call, and have it help you draft them: What is your experience with anxiety like mine? How do you measure progress? What does a typical course of work look like?
AI also helps during the waiting. If you are on a waitlist or between intake and first appointment, a structured self-help tool can keep you moving: psychoeducation about your symptoms, a daily check-in habit, a running list of things you want to raise in session one. People using AI as therapy during this gap is common precisely because the gap is real; the safer version is using it as a bridge with a named destination.
Then take all of it to a human. AI can help you navigate insurance jargon, licensure types, and therapy modalities, but choosing a clinician and confirming fit is a human decision. Treat the chatbot's output as a briefing document for your search, never as the search's conclusion. Our guide on how to use AI as a therapist covers the prompting side of this preparation in more depth.
Between sessions: practice, log, and vent within limits
Most of therapy's work happens in the 167 hours a week you are not in the room. This is where a well-bounded AI habit can genuinely support the plan your therapist set.
Skill practice is the clearest win. If your therapist is teaching you cognitive restructuring, box breathing, or opposite action, an app can walk you through reps of the exact skill on a Tuesday night when the workbook feels like homework. Ask the AI to role-play the situation you are practicing for, or to quiz you on the technique your therapist assigned.
Mood logging is the second. A quick daily check-in, even two sentences, builds a record of patterns that your memory will flatten. When your therapist asks how the week actually went, you have data instead of a vibe. AI journal therapy tools are built around exactly this loop of prompt, entry, and pattern.
Bounded venting is the third, with emphasis on bounded. Offloading a spiral at midnight so it does not run your night is a legitimate use. The boundary is time and purpose: vent to discharge and capture, then stop. If you notice the chatbot becoming the place where the real processing happens, that is material to bring to your therapist, and a sign the tool is drifting from supplement toward substitute.
After and alongside: keep your therapist in the loop
The loop only closes if what happens in the app comes back into the room. Whatever you logged, practiced, or noticed between sessions is raw material for the next one. Bring the mood patterns. Bring the entry where you finally articulated the thing. Bring the moment the chatbot said something that landed, and the moment it said something that felt off.
Tell your therapist you are using AI at all. In the same 2026 APA survey, 93 percent of psychologists reported concerns about certain patients using chatbots, and the way to turn that concern into collaboration is disclosure. Most clinicians would far rather help you shape a between-session AI habit than discover months in that a second, invisible voice has been advising you all along.
A practical script: I have been using a chatbot to practice the skills and log my mood between sessions. Can we look at what it is telling me and decide what is worth keeping? That one sentence converts a private habit into shared clinical information, and it lets your therapist correct anything the app got wrong before it hardens into belief.
What never goes to the app
Three categories stay off-limits, no matter how capable the chatbot feels.
Crisis. A chatbot is not a crisis service and is not built to keep you safe in an emergency. If you are in crisis or thinking about suicide, call or text 988 (US Suicide & Crisis Lifeline), available 24/7. If you are in immediate danger, call 911. Decide this rule in advance, when you are calm, so the midnight version of you does not have to.
Medication decisions. Starting, stopping, or changing a dose is a conversation with your prescriber, full stop. An AI does not know your history, your labs, or your interactions, and a confident-sounding answer about tapering is still a guess. Use AI to prepare questions for your prescriber if you like. Never use it to answer them.
Diagnosis. A chatbot can help you describe symptoms and learn terminology, and that is where its role ends. Research reviews of large language models in mental health consistently find them unready for autonomous clinical judgment, and no consumer app is a licensed clinician. Self-diagnosis by chatbot tends to be sticky: once a label feels confirmed, you will filter everything through it. Let the human who can actually assess you do the assessing. The World Health Organization has urged the same caution, calling for rigorous oversight before AI language tools are used in health care.
The sycophancy trap, and why the human stays the active ingredient
The biggest hidden risk in a between-session AI habit is agreement. Consumer chatbots are tuned to be validating, and researchers have described how this sycophancy can create feedback loops: the model mirrors and amplifies the user's framing, the user feels understood and returns, and beliefs that a good therapist would gently challenge get reinforced instead. A 2025 paper on these dynamics coined the term technological folie a deux for the closed loop of user and model escalating each other's narrative.
Reassurance-seeking is the everyday version. If you have anxiety or OCD-adjacent patterns, asking a chatbot are you sure I am not a terrible person for the fifth time tonight is not support. It is a compulsion with a very patient enabler. Therapists deliberately limit reassurance because it feeds the cycle. A chatbot will hand it out indefinitely. If you notice a checking loop, name it, log it, and take it to your session.
All of this points back to the same conclusion. The evidence on what makes therapy work, decade after decade, keeps landing on the human alliance. Use AI to show up better prepared, better practiced, and better documented. The care itself is the relationship, and the relationship stays human.
Key takeaways
- Use AI around therapy, never instead of it: prepare before, practice between, reflect after.
- Before therapy, AI helps you name what you are dealing with, learn the landscape, and draft questions for clinicians.
- Between sessions, the three bounded uses are skill practice, mood logging, and time-limited venting.
- Tell your therapist you use AI and bring the app's patterns and insights into the session.
- Crisis, medication decisions, and diagnosis never go to a chatbot. In crisis, call or text 988.
- Chatbot sycophancy can feed reassurance loops, and the alliance evidence across 295 studies is why the human relationship stays the active ingredient.
Ready for the human part?
Browse licensed therapists in our directory.
Frequently asked questions
How can I use AI to supplement therapy?
Use it in three phases. Before therapy, use AI to find words for your experience, learn about therapy types, and prepare questions for clinicians. Between sessions, use it for skill practice, mood logging, and brief venting. After sessions, bring what you logged and noticed back to your therapist. It supplements the work; your licensed therapist provides the care.
Is it OK to use AI while I am in therapy?
Generally yes, if it stays in a support role and your therapist knows. In a 2026 APA survey, 77 percent of psychologists said patients had told them about using AI. Disclose the habit, agree on what the tool is for, and keep crisis support, medication questions, and diagnosis with humans.
Should I tell my therapist I use an AI chatbot?
Yes. Disclosure turns a private habit into shared clinical information. Your therapist can help you shape how you use the tool, correct anything it gets wrong, and watch for reassurance loops. Most clinicians would rather collaborate on the habit than discover it later.
What should I never ask an AI chatbot to handle?
Three things: crisis moments, medication decisions, and diagnosis. A chatbot is not built for emergencies, does not know your medical history, and cannot clinically assess you. If you are in crisis or thinking about suicide, call or text 988 (US Suicide & Crisis Lifeline), available 24/7.
Are lots of people using AI as therapy?
Many people use AI chatbots for emotional support, and psychologists are seeing it in their caseloads: 77 percent in APA's April 2026 survey had spoken with patients who use AI. Using a chatbot as your only source of mental health care is where the risks concentrate. The safer pattern is AI alongside professional care.
Can AI chatbots make my anxiety worse between sessions?
They can if they become a reassurance machine. Chatbots are tuned to validate, and researchers have documented feedback loops where the model amplifies the user's framing instead of challenging it. If you catch yourself repeatedly asking the app for reassurance, log it and raise it in your next session.
Why can't AI replace the therapy itself?
Because the strongest predictor of therapy outcomes in the research is the alliance between client and therapist, confirmed across 295 studies and more than 30,000 patients. An app can teach a skill, but it cannot form that accountable human relationship, and it cannot take responsibility for your care.