When Therapy Meets AI: Who’s Really in the Conversation?
Author: Dr Vineetha Karuveettil1,2,3, Nivetha K4
1. JBI Amrita Centre for Evidence Synthesis and Implementation
2. Amrita School of Dentistry
3. Amrita Vishwa Vidyapeetham
4. Amrita Institute of Medical Sciences
In the movie Her, a lonely man begins talking to Samantha, an artificial intelligence (AI) designed to understand, respond and learn from him. What starts as conversation gradually becomes something much more personal.
When the film came out, the idea felt futuristic.
Today, asking AI about anxiety, loneliness or simply how to get through a difficult day doesn’t sound quite so futuristic.
That was something we kept coming back to while working together on our systematic review of AI-enhanced cognitive behavioural therapy (CBT) for depression and anxiety.
Can an algorithm actually help us feel better?
We went looking for the evidence
Our review began with a fairly straightforward question: How effective is AI-enhanced CBT compared with other non-pharmacological approaches for adults experiencing depression or anxiety?
What we discovered along the way was that ‘AI-enhanced therapy’ isn’t really one thing.
Sometimes it is a chatbot. Sometimes it is an app or online platform that personalises exercises, tracks emotions or provides feedback. Sometimes AI works in the background. And sometimes it works alongside a human therapist. Our forthcoming systematic review, currently under peer review, brought together 11 trials reflecting this diversity.
But we’ll spare you the forest plots.
The more interesting story was what we found behind them.
AI-supported CBT appears to offer some benefit for people experiencing depression and anxiety. But the effects were not consistent across every time point, and there is still considerable uncertainty in the evidence.
So, no – this isn’t a story about AI having cracked the code of mental health.
What interested us just as much was something much simpler: people seemed to stay engaged.
But what kept people coming back?
Across studies examining engagement and adherence, people using AI-supported programs generally participated more and dropped out less, although this wasn’t seen consistently across every study. Features such as reminders, personalised feedback and emotion tracking appeared to help people continue with the intervention.
And that made us wonder why.
Maybe sometimes it is easier to open an app at 11 pm than wait for the next appointment.
Maybe a reminder arrives on exactly the day you would otherwise have stopped.
Maybe there is comfort in being able to engage privately and at your own pace.
Our systematic review cannot answer those questions, and perhaps that is one of the most interesting things it revealed.
Where is the person in all this?
AI is moving extraordinarily quickly. The evidence supporting its use in mental healthcare is moving more cautiously.
Our review found encouraging signals, but also considerable uncertainty. Studies differed in what they called AI, how much human involvement remained and how interventions were delivered. Long-term evidence was particularly limited.
And there was something else we noticed.
Most of the evidence came from high-income countries.
As researchers working in India, this raised a line of questioning tied closely to home.
Would someone here engage with the same chatbot in the same way as someone in London, Sydney or California?
Would language matter? Culture? Digital literacy? Trust?
And perhaps most importantly: would people actually want AI involved in their mental healthcare – and if so, how?
We cannot answer those questions simply by looking at whether symptom scores improved in previous studies.
They require something AI cannot provide on its own: the voices of the people who actually use it.
Maybe Her asked the wrong question
The interesting question today isn’t whether humans can form a connection with AI.
People are already talking to technology.
For evidence-based healthcare, the harder question to answer is: when does that conversation actually help?
Our review suggests that AI-enhanced CBT has potential. But it also reminds us that evidence needs to keep pace with innovation. AI could support therapists, extend care beyond the clinic and help people remain engaged with treatment. At this stage, our findings point mainly towards AI complementing human care, rather than replacing it.
So, perhaps the future of mental healthcare isn’t human versus AI after all.
Perhaps it is about finding the right place for AI – while making sure that the person seeking help never becomes the least important part of the conversation.
References
Jesudason, D., Bacchi, S., & Bastiampillai, T. (2025). Artificial intelligence (AI) in psychotherapy: A challenging frontier. Australasian Psychiatry, 33(4), 629–632. https://doi.org/10.1177/10398562251346075
K, N., Karuveettil, V., Pillai, A. S., Jayachandran, M., & Pallaty, P. L. (2026). Effectiveness of artificial intelligence-enhanced cognitive behavioral therapy compared to alternative non-pharmacological therapies for treating depression or anxiety among adults: A systematic review protocol. JBI Evidence Synthesis, 24(1), 184–192. https://doi.org/10.11124/JBIES-24-00533
To link to this article - DOI: https://doi.org/10.70253/VJFW6482
Disclaimer
The views expressed in this World EBHC Day Blog, as well as any errors or omissions, are the sole responsibility of the author and do not represent the views of the World EBHC Day Steering Committee, Official Partners or Sponsors; nor does it imply endorsement by the aforementioned parties.
