← Kindness · 20 readings

Parenting Journal · Volume 2 · Issue 5 · Licensed source edition

Adolescents’ Well-being While Using a Mobile Artificial Intelligence–Powered Acceptance Commitment Therapy Tool: Evidence From a Longitudinal Study

Dana Vertsberger; Navot Naor; Mirène Winsberg; Khaled El Emam; Bradley Malin; Sebastian Fudickar; Zhenwei Dai

Originally published 2022-11-29 · JMIR AI
JMIR Publications Inc. · DOI 10.2196/38171 ↗

KindnessRetrospective longitudinal observational analysisCC BY 4.0
Ask this paper · Find source passages →

AI / system type

Automated ACT conversational service with human distress handoff; model architecture unspecified

Child-facing evidence

Examines children or child-focused systems. Transfer to everyday family life must be assessed separately.

The AI connection

Kai.ai offered automated ACT-oriented exercises and conversations in messaging apps, with a trained human companion taking over for detected distress. Architecture was not established as modern generative AI.

Why this reading matters

Combines a large apparent adolescent sample with unusually clear causal and age-verification limits, making it valuable for evidence literacy.

Editorial reading guide · Kindness

When a support chatbot reports that users feel better, how can I judge what caused the change?

Provides a useful lesson in interpreting impressive-scale emotional-support claims. Repeat respondents’ WHO-5 well-being scores rose during use, but the design cannot attribute improvement to Kai.ai. Distress could trigger a trained human companion, so the experience was not wholly automated. Parents can use this example to scrutinize what a companion’s evidence actually measures.

Study context

10,387 repeat questionnaire respondents selected from 43,237 Kai.ai users recruited through adolescent-targeted US social-media advertising.

Age / stage: Advertised to 14–18-year-olds; actual ages and gender were not collected or verified

Read with three questions

  1. How might selecting only repeat questionnaire respondents change the apparent trajectory of well-being?
  2. What alternative explanations remain without a control group, including regression to the mean and concurrent support?
  3. Why do unverified ages and handoff to a trained human companion limit claims about an autonomous adolescent AI therapist?

Keep the limits in view

No control group, voluntary repeat measurement and self-selection; improvement may reflect regression to the mean or other help. Actual age, gender and socioeconomic data were unavailable. Use of individual ACT/breathing features could not be tracked. Human distress support complicates attribution to automation.

The text below is the original article’s licensed repository edition. Its wording is not rewritten. Use the Paper Agent beside the text to locate evidence; open the source for figures, tables and complete context.

Opening the verified paper and agent…

Checking the article identity, license record, and saved source text.

Open the repository record

Attribution & reuse

Follow the original.

Original authorship, publication date, journal and DOI are preserved. This issue’s TRICK placement and reading questions are editorial additions, not claims made by the original authors.

CC BY 4.0 · License notice ↗Original repository record ↗

Dana Vertsberger; Navot Naor; Mirène Winsberg; Khaled El Emam; Bradley Malin; Sebastian Fudickar; Zhenwei Dai (2022). Adolescents’ Well-being While Using a Mobile Artificial Intelligence–Powered Acceptance Commitment Therapy Tool: Evidence From a Longitudinal Study JMIR AI. https://doi.org/10.2196/38171

Permissions checked against the saved article notice; repository publication notices screened 2026-11-04. This is not a new peer-review decision. The source authors and publishers are not represented as endorsing this issue.