Evaluating Generative AI Psychotherapy Chatbots Used by Youth: Cross-Sectional Study

Editorial reading guide · Respect

What happens to private information a teenager shares with an AI companion?

Accessibility and conversational fluency scored well, but understandable privacy policies, model transparency, therapeutic approach, and risk monitoring did not. This separates attractive interaction from respect for a young user’s information and safety needs.

Study context

Trained raters examined five widely used consumer chatbot platforms with the CAPE-II framework using conversations generated during August–October 2024.

Age / stage: Youth-targeted platforms; raters used simulated youth personas rather than child participants

Read with three questions

1. Which privacy and training-data disclosures were missing or hard to understand?
2. How did conversational quality differ from risk monitoring?
3. Why do persona audits tell us less about long-term youth outcomes than actual use studies would?

Keep the limits in view

The snapshot covered five entertainment/wellness platforms and a small set of simulated personas, with mostly single-session interactions and a 24-hour follow-up. It did not measure minors’ lived experiences, actual harms, or later platform versions.