Between sessions, engagement matters—but it is not the same as efficacy

The week between appointments

Therapy does not stop when the hour ends. Clients notice patterns, attempt exercises and encounter difficult moments in ordinary life. Yet a worksheet that made sense in session is easy to forget when the situation it addresses arrives three days later.

This is where carefully bounded conversational tools may be useful: not as independent therapists, but as a more accessible way to revisit psychoeducation, practice an agreed exercise, record a reflection or prepare something to discuss at the next appointment.

A strong signal for engagement

A 2025 randomized controlled trial followed 540 adults with elevated anxiety or depression symptoms. Participants received either a generative-AI CBT app or digital workbooks containing the same broad therapeutic curriculum. Over six weeks, the AI group opened the material 2.4 times more often and spent 3.8 times longer with it.

But anxiety and depression improved at comparable rates in both groups. The app was more engaging; it was not superior on the primary symptom outcomes. That distinction matters. A tool can make therapeutic material easier to return to without proving that conversation itself produces better clinical results.

More use is a pathway to care, not proof of better care. Clinics should track meaningful outcomes alongside opens, messages and minutes.

A second trial—and an important limitation

The 2025 Therabot trial randomized 210 adults with clinically significant depression, generalized anxiety or elevated eating-disorder risk to four weeks of a specialized generative-AI intervention or a waitlist. The intervention group showed larger symptom reductions, used the tool for more than six hours on average and reported a strong therapeutic alliance.

This was an important first clinical trial, but the comparison was a waitlist rather than an active digital or human treatment. Participants at high risk for suicide, mania, psychosis or a clinical eating disorder were excluded. The results support further study of purpose-built systems; they do not establish equivalence to a therapist or justify general-purpose chatbots for clinical care.

What therapist-guided continuity could look like

A clinic-aligned tool should begin with the care plan, not an open prompt box. The therapist defines appropriate exercises, tone, boundaries and escalation rules. The client knows what the system can do, what it cannot do and when a human will be contacted.

Useful signals then flow back into care without flooding the therapist: completed exercises, client-chosen reflections, outcome measures and clearly surfaced risk flags. The aim is continuity. The client receives support that remains recognizable as part of their therapy, and the therapist retains the context and authority to interpret it.