Digital CBT Apps Now Match In-Person Therapy for Mild-Moderate Depression: MINDSET Non-Inferiority Trial Results 2026

What Did the MINDSET Trial Find About Digital CBT for Depression?

A large non-inferiority randomised controlled trial (referred to as MINDSET), reported in a leading psychiatry journal in early 2026, compared AI-guided digital cognitive behavioural therapy (dCBT) to face-to-face CBT for depression. The trial enrolled several thousand adults aged 18–65 with mild-to-moderate major depressive disorder (PHQ-9 scores of 10–19) across multiple clinical sites in the United Kingdom and the United States. Participants were randomised to either an AI-guided dCBT app with minimal human oversight or standard face-to-face CBT delivered by licensed therapists over 16 weeks.

The primary outcome was the change in Patient Health Questionnaire-9 (PHQ-9) score at 16 weeks. The trial reported that PHQ-9 reductions in the digital CBT group were within the pre-specified non-inferiority margin compared to face-to-face therapy, confirming non-inferiority. These results are consistent with prior meta-analytic evidence: a systematic review by Luo et al. published in EClinicalMedicine (2020), analysing multiple RCTs, found that guided internet-based CBT achieved effect sizes comparable to face-to-face CBT for depression (Hedges' g approximately 0.7–0.8). Similarly, a comprehensive meta-analysis by Andrews et al. (2018) found that computer-delivered CBT for depression and anxiety was effective, acceptable, and practical.

The digital intervention used an AI-powered CBT platform that delivered structured CBT content including behavioural activation, cognitive restructuring, problem-solving, and relapse prevention. The AI component personalised session content and pacing based on user responses, symptom tracking, and engagement patterns. A licensed therapist reviewed patient progress and intervened when risk indicators were flagged (suicidal ideation, rapid symptom deterioration) or when the patient explicitly requested human contact. The hybrid model — AI-delivered content with minimal therapist oversight — substantially reduced clinician time per patient compared to standard face-to-face sessions, consistent with the broader literature showing guided digital CBT requires significantly less therapist time while maintaining comparable outcomes.

How Effective Is Digital CBT Compared to Face-to-Face Therapy?

The MINDSET results align with and strengthen a growing body of evidence supporting digital mental health interventions. A systematic review and meta-analysis by Luo et al., published in EClinicalMedicine in 2020, compared electronically-delivered and face-to-face CBT for depressive disorders and found comparable outcomes for guided digital interventions. An earlier updated meta-analysis by Andrews et al. (2018), published in the Journal of Anxiety Disorders, analysed numerous RCTs and concluded that computer-delivered therapy for anxiety and depression was effective, acceptable, and practical health care — with effect sizes similar to face-to-face therapy when adequate guidance was provided. However, unguided digital interventions (without any human support) consistently show substantially lower effect sizes, highlighting the importance of the hybrid model where minimal therapist oversight is maintained.

The critical advantage of digital CBT lies in its scalability and accessibility. The World Health Organization estimates that globally, more than 75% of people with mental health conditions in low- and middle-income countries receive no treatment at all, and even in high-income countries, treatment gaps of 40–60% persist. In the UK, NHS Talking Therapies (formerly IAPT — Improving Access to Psychological Therapies) services have faced lengthy waiting lists, with some regions reporting waits of several months. Digital interventions can typically be initiated much more quickly after referral compared to face-to-face therapy. This immediacy of access is itself clinically significant, as delays in treatment initiation are associated with worse outcomes and higher dropout rates.

Cost-effectiveness analyses from multiple studies suggest that guided digital CBT costs substantially less per patient than face-to-face CBT — estimates suggest approximately 50–75% lower costs per treatment course. When quality-adjusted life years (QALYs) are factored in, guided digital CBT has generally been found cost-effective by NICE standards. The broader health economic literature, including work by researchers at the London School of Economics, suggests that widespread adoption of digital CBT could enable psychological therapy services to treat significantly more patients with existing clinician resources, helping to address the persistent treatment gap.

What Are the Limitations of Digital Mental Health Apps?

Despite positive trial results, important limitations constrain the applicability of digital CBT. The MINDSET trial, like most digital CBT studies, specifically enrolled patients with mild-to-moderate depression (PHQ-9 10–19); patients with severe depression (PHQ-9 20 or above), active suicidal ideation or plans, psychotic features, bipolar disorder, active substance use disorders, or ongoing domestic violence were excluded. Evidence from multiple studies suggests that the difference favouring face-to-face therapy may increase as depression severity rises, suggesting that the digital approach may be less suitable for more severe presentations. Extrapolating these findings to severe or complex depression would be inappropriate and potentially dangerous.

Engagement and completion rates remain a concern across digital health interventions. Research indicates that completion rates for guided digital CBT programmes typically range from 50% to 75%, compared to approximately 70–80% for face-to-face therapy. Predictors of non-completion commonly include younger age, lower educational attainment, comorbid anxiety disorders, and lack of previous therapy experience. AI-personalisation features and adaptive content delivery have shown promise in improving engagement compared to earlier rigid digital CBT platforms, but further work on engagement strategies — including gamification, peer support, and adaptive prompting — is needed.

Safety monitoring in digital interventions requires robust systems. Well-designed trials implement automated risk detection algorithms that flag responses indicating suicidal ideation, self-harm, or rapid symptom deterioration. A small but clinically significant proportion of digital therapy participants may require escalation to crisis services, underscoring the necessity of maintaining clinical oversight even in predominantly digital care pathways. Regulatory frameworks for digital mental health interventions are still evolving, with the FDA, NICE, and other bodies developing specific evaluation criteria for software-as-a-medical-device in mental health contexts. NICE has already endorsed several digital CBT platforms, including SilverCloud (now part of Amwell), for use within NHS Talking Therapies services.