CBT homework: what an AI chat can do between sessions

Anyone who has been through CBT knows the arithmetic: the session lasts an hour, and the week has 168 of them. What happens in the other 167 is where therapy actually works.

That is why CBT sets homework: a thought record, graded exposure to something you have been avoiding, a breathing exercise, an activity to pick up again. It is not homework in the school sense. It is how a strategy stops being something you understood and becomes something you can do.

It is also the point where people drop out. Not out of laziness. The moment you need the exercise is almost always the moment you have forgotten you have it, or you do not feel like it, or it is eleven at night.

There is one finding worth knowing, and it is about exactly this gap.

In a randomised controlled trial published in Communications Medicine, 540 adults with symptoms of anxiety or depression were split into two groups for six weeks. One group used a cognitive behavioural therapy app built on generative AI, the other used standard digital workbooks.

The result: the app group came back 2.4 times more often and stayed 3.8 times longer. Symptom reduction and safety were comparable between the two groups.

In other words, people keep using it, while classic workbooks get abandoned sooner. For a piece of work that only helps if you actually do it, that is not a footnote. It is the main problem.

One more thing from the same trial: the people who used the personalised features saw a greater reduction in anxiety and better overall wellbeing.

The trial was not about psAico. It tells you something about the format, not about our app.

Remind you of the exercise at the moment you need it. A thought record works if you fill it in near the difficult moment, not three days later from memory.

Help you write it. “What went through your mind?” is a simple question that is hard to answer on your own. Having something that asks it, and that helps you tell the thought from the feeling, makes the exercise far less of an effort.

Explain again a technique you understood in the room and cannot recall now. It happens to everyone, and there is no sense waiting a week for something that takes three lines.

Keep track. Two weeks of ups and downs is worth far more in the session than “so so”.

Cover the gaps. Holidays, the session you had to miss, the stretch when your therapist is away and it would help simply to reread the exercises you were given.

There are no shades of grey here.

It must not give a second opinion on your therapy. If you find yourself asking a chat whether your therapist is doing the right thing, that question belongs to your therapist. It is awkward, and that is exactly why it is therapeutic.

It must not change your exercises. Graded exposure is graded by someone who knows you. An app that speeds it up because you sound motivated can do you harm.

It must not say anything about medication. Not whether to take it, not doses, not side effects. That is for your doctor.

It must not replace the session. If you start preferring the chat to the appointment, say so in the session. It is useful material, not a failing.

Tell your therapist. First and most important. Not to ask permission, but so that they know what you are working with between sessions.

Agree what it is for. “I use the app for thought records” is an agreement. “I use the app” is not.

Bring what comes out of it into the session. The material from the week, including the times you did not do the exercise. Especially those.

Keep it in proportion. It is a working tool, not a place to live. In psAico the four hour limit every seven days is there for exactly that reason.

This part is for people working clinically.

The most serious objection to a tool like this is that the client brings material into therapy that someone else produced. It is a fair objection, and the answer is not technical but clinical: the tool is agreed, the way any homework is agreed, and what comes out of it is brought into the session like any other material.

The finding from the trial with 540 people is worth holding on to: the demonstrated advantage is on consistency. That is the part of the work clients drop first, and the part a therapist cannot supervise from a distance.

psAico grew out of the clinical work of the psychology practice of Dr Michele Facci in Italy, and it is built to sit in that space: psychoeducation, self monitoring, support with exercises. It does not make diagnoses, it does not comment on which professional a person should see and it does not touch medication, and the evidence behind it is set out in full.

• McFadyen J, Habicht J, Dina LM, Harper R, Hauser TU, Rollwage M. Increasing engagement with cognitive-behavioral therapy (CBT) using generative AI: a randomized controlled trial (RCT). Commun Med (Lond). 2026 Jan 15;6(1):129. doi: 10.1038/s43856-025-01321-8. PMID: 41540194.
• Sohn JS, Ha BG, Park S, Kim JJ, Lee E, Oh H, Lee S, Kim E. Systematic review and meta analysis of chatbots in the management of depressive and anxiety symptoms. NPJ Digit Med. 2026 Mar 25;9(1):377. doi: 10.1038/s41746-026-02566-w. PMID: 41882250.