A youth chatbot can be feasible without its effect being proven yet

The pilot took place in primary care

Adolescents between 13 and 17 years with moderate depressive symptoms who were treated in an academically affiliated network of primary care practices could participate.

Ten individuals were assigned to the app and eight to a waitlist. One individual proved ineligible upon record review, so 17 were included in the analysis.

The primary endpoint was the PHQ-9 after four weeks. Additionally, acceptance, feasibility, and usability were assessed from the perspective of the adolescents and their parents or guardians.

Interviews with 13 professionals from eleven clinics aimed to reveal barriers and facilitating conditions for integration into treatment plans.

The sample was very small and lacked diversity

Of the 17 adolescents evaluated, 15 were female, white, and privately insured. This composition limits the generalizability to the broader adolescent population.

Digital care in particular is often justified by better access for underserved groups. A predominantly privileged sample cannot yet test this claim.

The authors therefore call for larger studies involving rural, socioeconomically disadvantaged, and underrepresented communities. Such groups must not only be recruited but also included in design and implementation.

Language, device access, family support, and local care provision also influence whether an app is practically usable. Scalable software does not automatically create equal access.

The PHQ-9 trajectories provide only an initial estimate

After four weeks, the mean PHQ-9 score in the app group decreased by 3.3 points. This shifted the group average from the moderate to the mild symptom category.

In the waitlist group, the score decreased by two points, without a shift in the mean category. The difference is interesting as a preliminary signal, but highly uncertain given ten and eight randomized participants, respectively.

A small sample cannot reliably compensate for random group differences and can barely capture rare problems. An observed mean must therefore not be treated as a stable effect size.

The pilot was intended to provide feasibility and estimates for subsequent research. Measuring it against a definitive efficacy benchmark would be just as wrong as marketing it as evidence of efficacy.

Acceptance and usability were high

Adolescents and parents reported high usability, feasibility, and acceptance. This is essential for planning a larger study: a service that is meaningful in content is of no use if it is not adopted in everyday life.

Acceptance can be fostered by familiar mobile use, immediate access, and the conversational format. However, it does not automatically indicate whether individual responses are professionally or emotionally appropriate.

For a comprehensive assessment, qualitative conversational errors should be documented: rigid repetitions, unsolicited advice, incomprehensible exercises, or difficulties after a correction.

For minors, consent, parental roles, and privacy protection are also part of the user experience. The service must explain what remains confidential and when other levels of care are involved.

The high acceptance in a supervised pilot project can also benefit from introduction and contact persons. A freely accessible service without a clinical framework must separately examine whether adolescents understand functional limits, data use, and next steps equally well.

Professionals preferred early supplementary use

The surveyed primary care providers preferred using such mHealth offerings early in the course of treatment for mild or moderate symptoms.

This assessment positions the app as a supplement within an existing network, not as an isolated replacement for any form of care. Professionals can guide introduction, expectations, and subsequent steps.

Integration requires clear feedback structures. Which usage outcomes are relevant for practice, who sees them, and how is it prevented that sensitive conversation data unnecessarily flows into records or systems?

Organizational fit is part of evidence of efficacy in everyday life. An app can function technically and still fail if no one knows how its information can be responsibly incorporated into care.

The early Woebot study shows a different age window

Fitzpatrick, Darcy, and Vierhile randomized 70 young adults aged 18 to 28 years to two weeks of Woebot or an informational e-book. That was a larger but also short and unblinded study.

In the intention-to-treat analysis, the Woebot group reduced their PHQ-9 scores more. Among those with complete data, anxiety scores decreased in both groups.

Participants used Woebot an average of 12.14 times. Their comments suggested that process factors were more important for acceptance than traditional therapeutic content alone.

The comparison shows that findings from older adolescents and young adults cannot simply be transferred to 13-year-olds. Developmental stage and care context alter the requirements.

The meta-analysis finds a small effect on distress

Li and colleagues screened studies from 2014 to September 2024 and included 29 interventions for young people, among them 13 randomized controlled trials.

In the meta-analysis, chatbot interventions reduced psychological distress with a small effect size of Hedges g minus 0.28. No significant effect emerged for psychological well-being.

The effect varied by sample, platform, mode of interaction, and type of response generation. A chatbot is therefore not a uniform intervention but a surface for very different contents and systems.

Overall, feasibility and acceptance were rated positively. The authors see potential as a complement to multidisciplinary care and identify room for improvement in language, accuracy, data protection, and safety.

Adolescents require their own evidence pathway

The pilot shows that a CBT app with a conversational agent was acceptable and practically usable in a small primary care sample. It cannot definitively determine efficacy.

Subsequent experiments must be larger, more diverse, and longer. They should examine not only scales, but also usage, dropout, conversation quality, adverse effects, and integration into real-world care.

For an offering aimed at ages 16 or 18, age limits are not a purely marketing decision. Content, agreement, data protection, language, and lines of responsibility must match the target group.

A youth chatbot can be feasible without its efficacy having been proven yet. It is precisely this honest intermediate stage that enables good research: first test whether the system can be deployed sustainably, then rigorously investigate whom it helps under what conditions.