AI & Mental Health · 4 of 9
Bring the Transcript. It Tells Me Things Three Sessions Wouldn't.
How to use AI as a genuine adjunct to therapy, from a psychologist who'd rather you used it well than pretended you weren't.
How do you use AI alongside therapy? Use it for the tasks that involve organising your own thinking: capturing what happened between sessions, drafting difficult conversations, understanding terminology, preparing what you want to raise. Stop where the work needs judgement about you. Tell your clinician you're using it.
"I'd far rather a patient used a chatbot openly and well than hid it and used it badly. The hiding costs us more than the tool ever does."
A patient asked me last year whether she should delete the app. She'd been using it between sessions and felt guilty about it, as though she'd been unfaithful to the process.
I asked to see what she'd written. It was three months of careful notes about her own week, more detailed than anything she'd managed to tell me in the room, because she'd written them on the day rather than reconstructing them a fortnight later. It was, in effect, exactly the between-session record I ask people to keep and almost nobody keeps.
What it's genuinely good for
The tasks where the value comes from you doing the thinking and the machine handling the structure.
- Capturing the week. Memory for emotional detail decays fast. Writing on the night it happened gives you something accurate to bring, instead of a summary shaped by how you feel today.
- Drafting the difficult conversation. What you'll say to your mother, your manager, your spouse. Rehearsal is legitimate work and AI is good at it.
- Understanding the vocabulary. What a term means, what a therapy involves, what a medication class does in general. Information delivery is the thing these systems are built for.
- Preparing for a session. Working out what you actually want to raise, so the first ten minutes aren't spent finding it.
- The middle of the night. When the choice is writing to something or lying awake, writing is better.
- Practising a skill. If your clinician has given you a technique, having something walk you through the steps between sessions is reasonable.
What clinicians are seeing
Sources: American Psychological Association practitioner survey, April 2026; Mental Health UK / Censuswide polling, November 2025; India Global Analysis of Policy report on chatbot downloads, January 2026.
Where to stop
The line sits at judgement. Anything that requires an opinion about you rather than help arranging your own material belongs with a person.
Don't ask it who was right
This is where sycophancy does its damage. It will side with you, because you're the one describing the situation. Ask it instead to write the other person's version as sympathetically as it can, which is uncomfortable and far more useful. The mechanism is set out in why AI agrees with you.
Don't ask it what's wrong with you
It will produce something plausible from the words you gave it. Diagnosis needs duration, history, functioning, physical causes and differentiation between conditions that look alike. A chatbot's answer is a reason to get assessed, not the assessment.
Don't use it at your lowest
Safeguards weaken across long conversations, and UK polling found a small but real proportion of users reported harmful material. Your worst night is the night to reach a person.
Don't let it become the only place
Watch for the moment your family and friends start getting an edited account while the full version goes somewhere else. That's the substitution point. The other signals are here.
Conversations with general-purpose chatbots are not confidential in the way clinical sessions are. They are typically stored and may be used to improve the system. Assume anything you type exists somewhere, and be deliberate about names and identifying details, particularly about other people.
A working method
What I suggest to patients who are already using it.
- Write on the day, not the week after. Short and specific. What happened, what you felt, what you did. You're building a record, not an essay.
- Ask for the counter-case deliberately. "Give me the strongest argument that I'm the difficult one here." Then read it properly instead of dismissing it.
- Turn it into three lines before a session. The three things you most want to say. Bring them written down.
- Keep the transcript. Or the parts that matter. It's a record of your own thinking over time, which is genuinely hard to get otherwise.
- Tell your clinician. The tool isn't the problem. Not knowing about it is.
Why I want to know
Because of what it shows me. How you narrate yourself when you think no one's listening is different from how you narrate yourself to me in the first month, and the gap between those two accounts is often the most useful thing in the room.
It also shows me where you got stuck. If you've been circling the same point for eleven weeks, that point is where we should start, and without the record I might take three sessions to find it.
This matters especially for people working with me from outside India, where sessions are often further apart and a lot of life happens between them. A written record of the gap is worth more there than anywhere.
Bring what you've written
Sessions with Dr. Kohli directly, in Gurugram or online worldwide, in English or Hindi.
Book a session WhatsApp +91 9811862338The AI and mental health series
Start with the overview, or read across the set.
- AI and your mental health — the overview.
- Is ChatGPT a good therapist?
- Why AI agrees with you
- When to see a person, not an app
- When your teenager confides in a chatbot
- Using AI alongside therapy (you're here)
- Is ChatGPT confidential?
- AI told me I have depression
- AI, distance and the diaspora
Common questions
Can I use ChatGPT between therapy sessions?
Yes, and it can genuinely help if you use it for the right tasks. Capturing what happened while it's fresh, preparing what you want to raise, and practising a technique your clinician has given you all support the work. Asking it to judge situations or diagnose you competes with the work instead.
Should I tell my therapist I use AI?
Yes. Most clinicians find it useful rather than offensive, and hiding it means we're working with less information than you have. It also tends to come out eventually, at which point the concealment has become its own subject.
Will my therapist be offended?
A good one won't. You're doing something between sessions to manage your own difficulty, which is what we ask people to do. The specific tool matters less than whether it's supporting the work or quietly replacing it.
What should I actually type?
Concrete accounts rather than requests for verdicts. "This is what happened on Tuesday and here's what I felt" produces something useful. "Was I wrong?" produces agreement. The first gives you material. The second gives you reassurance you didn't need.
Is it safe to share personal details with a chatbot?
Be deliberate. These conversations are typically stored and may be used to improve the system, and they aren't confidential in the clinical sense. Writing about your feelings is one thing. Naming other people, workplaces or specific circumstances is worth thinking twice about.
Can AI help me practise CBT techniques?
It can walk you through the steps accurately, which has value if a clinician has already established that the technique is right for you. What it can't judge is whether you're doing the exercise while avoiding the actual difficulty, which is a common and invisible problem.
How do I stop AI from just agreeing with me?
Ask directly for the opposite. "Write my husband's account of this argument as he would tell it." "What would change your view of who's at fault?" The model follows explicit instructions well. What it won't do is volunteer disagreement, which is when you most need it.
Should I bring transcripts to a session?
If you'd like to. Several people have and it's shortened the early work considerably. Bringing the parts that matter is usually better than the whole thing. What's most useful is where you kept returning, not the volume.
Can AI replace journalling?
It's a form of journalling with a response attached, and for many people the response is what makes them keep going. The risk is that the reply becomes the point and you stop reading your own writing. Try reading back a week of what you wrote without looking at the answers.
Is it a problem if I use it every day?
Frequency isn't the measure. Whether it's the only place you speak honestly is. Daily use alongside real conversations and a clinician is fine. Daily use that has replaced telling anyone in your life how you are has moved into different territory.
What if AI contradicts my therapist?
Bring it in. That's a genuinely useful conversation, and a clinician who can't discuss why they take a different view isn't giving you much. The one thing worth remembering is that the model doesn't know you and has heard only your account of what was said.
Can I use AI for couples issues if my partner won't come to therapy?
Carefully. It's useful for drafting how you'll raise something and for generating your partner's perspective if you ask for it explicitly. It's poor at judging the relationship, because it only ever hears one side. Relationship work can start with one person in the room, and often does.
Related reading: anxiety, depression, and marriage counselling.
This article is general psychoeducation and not a substitute for individual assessment or treatment. If you are in immediate danger or unable to keep yourself safe, contact your local emergency services or attend your nearest hospital without delay.