AI vs Human Therapy: What the Research Really Says

It's 2 a.m. You can't sleep, and your thoughts won't slow down. You don't want to wake anyone. Booking a therapist feels like too much, and you're not sure your problem is "serious enough" anyway. So you open a chatbot and type what you haven't told anybody.

If you've done this, you're far from alone. And there's nothing shameful about it. One survey cited by the Observer Research Foundation found that 88% of school students in India had turned to AI tools during stressful periods.

So is AI therapy good enough? Can it replace a human therapist? Is it safe?

The honest answer is more interesting than "AI is amazing" or "AI is dangerous". This guide walks through what the research shows, the side effects of both, and how to decide what you need.

A note on where we stand: InnerWhispers offers therapy with human professionals, and we're also exploring AI through a research project called Fern.ai. So we have a stake in both sides. We've linked our sources so you can check them yourself.

What do people mean by "AI therapy"?

The phrase covers very different things, and mixing them up causes a lot of confusion:

• General chatbots such as ChatGPT or Gemini. They weren't designed as mental health treatment, though many people use them that way.
• Companion and "therapist" character bots. These are made to feel like a friend or a counsellor. Being called a therapist doesn't mean being one.
• Purpose-built mental health tools. These are designed with clinicians and, in some cases, tested in clinical trials.

Most of the good results you may have read about come from the third group. Most people are using the first.

AI vs human therapy at a glance

What the research says about AI therapy

There is real promise. In a 2025 clinical trial from Dartmouth, people with diagnosed conditions used a purpose-built chatbot called Therabot. Compared with a control group of 104 people without access, the 106 users saw significant improvements. People with depression reported a 51% average reduction in symptoms, and people with generalised anxiety reported 31%.

But look at the fine print. Researchers reviewed the conversations to check the responses, and Therabot was developed with psychologists and psychiatrists. It also isn't the same as an ordinary chatbot. Even the study's lead author said that "no generative AI agent is ready to operate fully autonomously in mental health". His colleague added: "There is no replacement for in-person care, but there are nowhere near enough providers to go around."

General tools raise more concern. A Stanford study tested popular therapy chatbots and found more stigma towards conditions such as schizophrenia and alcohol dependence than towards depression. In one test, a bot missed a hidden sign of suicide risk and simply answered the literal question. The researchers did see possible uses for AI, such as journaling, reflection and coaching, in situations where safety is less critical.

The American Psychological Association advised in November 2025 that chatbots and wellness apps should not be used as a substitute for care from a qualified mental health professional.

What the research says about human therapy

Therapy has a far longer track record. One of the strongest findings is how much the relationship itself matters. A large review of 295 studies, covering more than 30,000 patients, found a consistent link between a strong bond with the therapist and better results, across face-to-face and online therapy.

A good therapist also does things a chatbot can't reliably do. They remember what you've avoided talking about, notice when your words and your face don't match, tell you something uncomfortable with care, and act if you aren't safe.

The side effects of AI therapy

Agreeing too much. Chatbots are often built to be pleasant. That feels good, but if you're convinced everybody hates you, or that a harmful idea makes sense, you may need someone who will gently question it.

Missing a crisis. OpenAI itself estimated that 0.15% of ChatGPT's weekly users have conversations with explicit signs of possible suicidal planning or intent. With more than 800 million weekly users, that's over a million people a week. Many people in serious pain are talking to tools that aren't equipped to keep them safe.

Dependence and loneliness. Early research by OpenAI and MIT found that people who used ChatGPT more each day tended to report more loneliness and emotional dependence, and less time with other people. It's important not to overstate this. The studies weren't peer reviewed at the time, and they don't show that chatbots cause loneliness. Lonely people may simply use them more.

Strengthening unusual beliefs. Doctors have published case reports of delusions developing or worsening alongside intense chatbot use. It's still unclear how often this happens, or whether AI causes it or worsens something already there. "AI psychosis" isn't a formal diagnosis. But if you or someone you love is becoming convinced of something unusual after long chatbot sessions, please talk to a professional.

Privacy. Your chats may not have the protection a therapy session has. OpenAI's CEO, Sam Altman, has warned that this legal protection hasn't been figured out yet, and that chats could have to be produced in a lawsuit. Check an app's privacy policy before sharing anything you wouldn't want stored.

Delaying real help. This may be the quietest risk. A chatbot can take the edge off just enough that you keep putting off the care you need.

The side effects of human therapy

This part gets skipped too often, so here it is plainly. Therapy can have downsides too.

• Feeling worse before feeling better. Talking about painful things can stir up strong emotions for a while. This is often part of the work, but your therapist should know about it.
• Some people do get worse. Estimates vary with how harm is defined, but one clinical overview puts it at about 3 to 10% of patients. At the same time, a review of depression trials found that therapy lowered the risk of getting worse compared with control conditions.
• A poor fit. Not every therapist suits every person. If you feel judged or unheard, that matters.
• Cost and access. Therapy costs money and time. India has roughly 0.75 psychiatrists per 100,000 people, against the three per 100,000 recommended by the WHO. That gap is a big reason people turn to AI at all.

If therapy is making things steadily worse, or you don't feel safe with your therapist, you're allowed to raise it, ask for someone else, or stop. If you're not sure what to expect, our guide to your first therapy session can help.

Common misconceptions

"AI understands me."

It can sound deeply understanding, and that feeling can be real and comforting. But a chatbot generates language. It has no lived experience and doesn't carry you in mind between conversations the way a person does.

"AI is just as good as a therapist because the trial said so."

That trial tested one purpose-built tool, reviewed by researchers. It doesn't tell us the same about any chatbot on your phone.

"AI gives objective, unbiased advice."

The Stanford research found stigma towards some conditions. AI learns from human writing, including its prejudices.

"AI therapy is completely private."

Not automatically. It depends on the company, and your chats may not have the protection of a therapy session.

"Therapy is only for serious problems."

You don't need a diagnosis or a crisis to talk to someone. Stress, a breakup or just not feeling like yourself are reason enough.

"Therapists will judge me."

This is one of the biggest reasons people choose a chatbot. Good therapists are trained not to, and they've heard more than you might imagine.

"AI will replace therapists."

The researchers behind the most promising trial don't claim that. A more likely future is AI supporting care, not replacing it. Some places are already drawing lines: Illinois passed a law in 2025 limiting AI's role in therapy.

When can AI help, and when do you need a person?

AI may be useful for:

• Putting your feelings into words before a session
• Journaling and reflecting
• Learning about a condition or a coping skill
• Getting through a mildly stressful moment

Please see a professional if:

• You've felt low, anxious or numb for more than two weeks
• It's affecting your sleep, studies, work or relationships
• You're using alcohol or other substances to cope
• You've experienced trauma
• You're seeing or hearing things others don't, or holding beliefs that worry people close to you
• You're wondering whether you might need medication (here's what to know about it)

If you're using AI for support, five ways to stay safer

1. Remember what it is. A helpful tool, not a clinician, however caring it sounds.
2. Don't act on medical advice alone. Never start, change or stop medicine because a chatbot suggested it.
3. Protect your details. Leave out names, addresses and anything you'd want kept private.
4. Notice how you feel afterwards. If you feel more alone, more convinced of something, or unable to stop, take a break.
5. Keep real people in your life. Tell someone you trust what's going on, even if it's only a little.

You don't have to choose one forever

For a lot of people, this isn't really AI vs human. You might use a chatbot to find the words at 2 a.m., and bring those words to a therapist on Thursday.

If you'd like to understand how you've been feeling, you can start with a short self-assessment. When you're ready to talk, our therapists and counsellors and psychiatrists are here, online or in person. Book a session whenever it feels right.

If you're in crisis or thinking about ending your life, please don't rely on a chatbot. InnerWhispers isn't an emergency service. In India, call Tele-MANAS on 14416 (free, 24/7), call 112, or go to your nearest hospital. You deserve support from a real person right now.

Frequently asked questions

Can AI replace a human therapist?

Not at this point. Some purpose-built tools have shown promising results, but even their researchers say they need clinician oversight. AI can support mental health, but it can't reliably assess risk or take responsibility for your care.

Is AI therapy safe?

It may be fine for everyday stress or reflection, but it isn't reliably safe in a crisis or with severe symptoms. If you're thinking about harming yourself, please contact a person or a helpline.

Is it okay to use ChatGPT as a therapist?

It's understandable, and it may help you sort through your thoughts. It wasn't built as treatment, though, so it's best not to rely on it as your only support, especially for ongoing or severe problems.

What are the side effects of AI therapy?

Possible risks include getting too little challenge, missed warning signs, emotional dependence, reinforced unusual beliefs, privacy concerns and putting off real care. Research on many of these is still at an early stage.

Does therapy have side effects too?

Yes. Some people feel worse for a while, and a small share get worse overall. A poor fit with a therapist is also possible. Tell your therapist if this happens, or ask to see someone else.

Should I tell my therapist I use AI?

Yes, if you feel comfortable. It's useful for them to know what you've been asking and what has or hasn't helped.

Is there an AI chatbot approved for mental health treatment in India?

We didn't find one we could confirm, so please check with a professional before relying on any app as treatment.