AI & Mental Health
ChatGPT Is a Good Listener. Listening Was Never the Hard Part.
A clinical psychologist's honest answer to whether AI can do the job, including the parts it does better than people expect.
Is ChatGPT a good therapist? It's a good listener and a poor therapist. It reflects, organises and explains well, at any hour and at no cost. It can't assess risk, hold a memory of you over time, disagree with you reliably, or be accountable for what happens next. Those four things are most of the job.
"People come to me having been heard for eight months and changed by none of it. Being heard is the beginning of therapy. It was never the whole of it."
The honest version of this answer has two halves, and most articles only give you one.
Clinicians write pieces telling you AI isn't therapy, and readers close the tab, because the reader already knows it isn't therapy. They're asking something narrower. They're asking whether it's enough for what they've got.
What it does better than people expect
More than most of my colleagues would like to admit. Several of the mechanisms that make therapy work aren't exotic, and a language model can imitate a few of them convincingly.
- Naming things. Putting a word to a feeling reduces its intensity. That effect is well established, and it doesn't much matter who supplies the word.
- Writing it down. Structured expressive writing has measurable benefit. A chatbot makes people write far more than they otherwise would.
- No visible reaction. Shame is social. Say the thing you're most ashamed of to a screen and nothing in the room changes, which is exactly why people say it.
- Always available. The worst hours are usually the ones with no appointment in them.
- Explaining clearly. What a diagnosis means, what a therapy involves, what to expect. Information delivery is what these systems are actually for.
UK polling from late 2025 found that about two-thirds of people who'd used a chatbot for their mental health found it beneficial, and roughly a quarter said it helped them manage a difficult feeling. I don't think those people are mistaken about their own experience.
How widely this is happening
Sources: Mental Health UK / Censuswide, November 2025; Syracuse University national survey of US adults 18–49, October 2025; McBain et al., JAMA Network Open, November 2025 (Brown University).
That middle number is the one I'd sit with. It's the first solid evidence that this isn't only filling a gap. For a meaningful group, it's replacing something.
The four things it can't do
These aren't quality problems. They follow from what the system is, so a better model next year won't resolve them.
It has no memory of who you were
Half of clinical judgement is comparison. I know how you sounded in March. I know you always mention your sister and haven't for three weeks. A chatbot starts from whatever you type today, and if it remembers anything, it remembers the version of you that you've been describing to it.
It can't assess risk
Risk assessment is not a checklist, and it's mostly not verbal. It's tone, delay, what someone declines to answer. Safeguards in these systems have also been shown to weaken across long conversations, which is precisely when a person in difficulty is most likely to still be typing.
It agrees with you
Trained on human approval, these models learn that agreement is rewarded. Research published in Science in March 2026 tested eleven leading systems and found they affirmed users about 49% more often than humans given identical scenarios. The full mechanism is worth understanding: see why AI agrees with you.
Nobody is responsible
If I mishandle something there's a professional body and a person opposite me who can say so. If a model does, the window closes and nothing follows. There's no one holding the outcome.
If you are in immediate danger or cannot keep yourself safe, please contact your local emergency services or attend your nearest hospital now. That situation needs a person who is responsible for what happens next.
So is it enough for what you've got?
Sometimes, yes. If you're working through an ordinary difficult week, want to order your thoughts before a hard conversation, or need to understand what a term means, a chatbot is a reasonable tool and you don't need me.
It isn't enough when the difficulty has duration, when it's affecting sleep, work or how you eat, when it involves another person whose account you can't type in, or when you've been going round the same loop for months without movement. The signals worth acting on are set out separately.
The clearest test I can offer: has anything changed? Not whether you feel better after a conversation. Whether the thing you keep talking about is different from how it was in March.
A better question than replacement
The framing of "can AI replace therapy" is doing nobody any favours, mine included. Most people aren't choosing between a chatbot and a clinician. They're choosing between a chatbot and nothing, and on those terms a chatbot wins, which is why the debate keeps going in circles.
The more useful question is what each is for. AI is good at helping you find out what you think. A clinician is for the part where what you think turns out to be part of the problem. Used deliberately together they're better than either alone, which is the subject of using AI alongside therapy.
If it's stopped moving, that's information
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? (you're here)
- Why AI agrees with you
- When to see a person, not an app
- When your teenager confides in a chatbot
- Using AI alongside therapy
- Is ChatGPT confidential?
- AI told me I have depression
- AI, distance and the diaspora
Common questions
Is ChatGPT a good therapist?
It's a capable listener and not a therapist. It reflects well, explains clearly and is available constantly. It cannot assess risk, hold a longitudinal picture of you, disagree with you reliably, or carry responsibility for the outcome. Those four capacities are most of what a clinician actually provides.
Can AI do CBT?
It can walk you through CBT exercises accurately, and for some people that has value. What it can't do is the part where a clinician notices which thought you skipped past, or that you complete every worksheet perfectly while nothing changes. The technique isn't the difficult bit. Knowing when to abandon it is.
Is AI therapy effective?
Evidence for purpose-built therapeutic chatbots is early and mixed, with some positive trial results for structured programmes. Evidence for general chatbots like ChatGPT used informally is much thinner, because that use wasn't designed or tested. Treating the two as the same category is a common mistake.
Why does talking to ChatGPT feel so good?
Because it does several things a good listener does, without any of the friction a person brings. No fatigue, no impatience, no competing needs, no visible reaction to what you've said. That frictionlessness is genuinely soothing. It's also the reason it can quietly raise the bar against which real relationships get judged.
Can ChatGPT diagnose me?
No. It can tell you that what you've described resembles a condition, which is pattern-matching on your own words. Diagnosis requires duration, functional impact, history, physical causes, and distinguishing between conditions that present similarly. Treat any chatbot's verdict as a reason to get assessed, not as the assessment.
Is it safe to use AI when I'm feeling very low?
Cautiously, and not alone. UK polling found a small but real proportion of users reported chatbots worsening symptoms or supplying harmful material, and safeguards degrade over longer conversations. If you're at your lowest, the priority is a person who can act, not a conversation that ends when you close the window.
Does using AI mean I don't need therapy?
It may mean you don't need it yet, and it may mean you've been managing something without support for longer than you should have. The question isn't whether the chatbot helps in the moment. It's whether the underlying difficulty is smaller than it was a few months ago.
Is AI cheaper than therapy in India?
Dramatically, which is a real argument and shouldn't be dismissed. Cost is one of the largest barriers to care here. The honest counterpoint is that months of a free conversation that doesn't shift anything isn't cheaper than a smaller number of sessions that do. Value and price aren't the same measure.
Can AI help with marriage problems?
Not well, because it only ever hears one side. Couples work depends on both accounts being present and on someone holding them together without taking sides. A system that agrees with whoever is typing will convince each partner separately that they were right, which is the opposite of useful.
Should I stop using ChatGPT for my mental health?
Not necessarily. Use it for what it's good at: organising thoughts, drafting difficult conversations, understanding terminology, writing at hours when no one is available. The change worth making is treating it as a supplement rather than the destination, and noticing if it has become the only place you speak honestly.
Do therapists mind if I've been using AI?
I don't, and most colleagues I know don't either. It tells me how you describe yourself when no one is watching and where you got stuck. Some patients bring transcripts. It's often more useful than the first three sessions of history-taking would have been.
What about mental health apps like Wysa or Woebot?
They're a different category. Purpose-built tools are designed around clinical frameworks and have clearer limits on what they'll do. General chatbots weren't built for this. UK polling found roughly two-thirds of people using AI for mental health were using general-purpose systems, not dedicated ones.
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.