Psychology› Applied Psychology› Digital Mental Health Interventions

Digital mental health — Fido, a Polish-language CBT therapy chatbot

Submitted 2026-10-22

Field: Psychology › Applied PsychologyType: Couldn't replicate a published resultArea: digital mental healthTarget: Fido, a Polish-language CBT therapy chatbotStage: randomized trial

Hypothesis

Two weeks with a fully automated CBT chatbot would reduce depressive and anxiety symptoms more than self-help material, as reported for Woebot in English.

What failed

Among 81 young adults with subclinical symptoms (36 chatbot, 38 control), symptoms fell in both arms and stayed down at one-month follow-up with no difference between them; loneliness also did not differ, and control participants in fact spent longer on their intervention (mean 118 against 79 minutes).

Broken assumption

A guided chatbot outperforms a self-help book rather than merely matching it.

Why

The authors state they did not replicate the earlier chatbot findings, because the comparison condition improved as much as the chatbot rather than because the chatbot failed outright; a frequency-of-use analysis of loneliness was exploratory only.

From the published paper: Karkosz et al. JMIR Formative Research, 2023. ↗CC BY

Research ideas from this failure

Unvetted lead — read the papers

New hypotheses this failure suggests, each with a crucial experiment, a rival explanation of the original result, and the papers it rests on. Written by AI from paper titles and abstracts found on OpenAlex (and Europe PMC, for life and health sciences).

Related failures (1)

Matched on shared genes, mechanisms and ways of failing as well as wording. Each shows why it matched — treat these as leads rather than verdicts.

digital mental healthMachine-learning dropout prediction on small datasets
Machine-learning models trained on the small datasets typical of digital mental-health studies give trustworthy estimates of how well dropout can be predicted.
Learning curves on 3,654 users of one digital intervention showed that datasets of N ≤ 300 overestimated predictive power; for uninformative features, in-sample performance even fell as datasets grew, and sophisticated models overfitted small samples. Overfitting eased at N = 500, but performance did not converge until N = 750–1,500.
Why it matched: same research area

Links in published research

Where this experiment's genes are known to matter outside its own research area, and the papers behind those links. Sources: Open Targets Platform and Europe PMC.