Truace
Subscribe
70/100

Moderate evidence

Cite
More
Export CSVExport JSONEmailLinkedInX

Chatbots modestly cut distress in young people, a review of 11 trials finds

Study title: Chatbot‐Delivered Interventions for Improving Mental Health Among Young People: A Systematic Review and Meta‐Analysis

Evidence type
Systematic review or meta-analysis
Sources
1 sourcePeer-reviewed
Published
3 min read
Truvace record
TRV-2026-2736cite or check it

Study title: Chatbot‐Delivered Interventions for Improving Mental Health Among Young People: A Systematic Review and Meta‐Analysis

A systematic review and meta-analysis in Worldviews on Evidence-Based Nursing pooled randomized trials of chatbot-delivered mental health interventions for people aged 10 to 24. Across 11 trials, chatbots reduced psychological distress (Hedges' g = -0.28, 95% CI -0.46 to -0.10). Across seven trials, they had no significant effect on well-being (g = 0.13, 95% CI -0.16 to 0.41).

Read the study abstract

BACKGROUND: The characteristics, application, and effectiveness of chatbots in improving the mental health of young people have yet to be confirmed through systematic review and meta-analysis. AIM: This systematic review aims to evaluate the effectiveness of chatbot-delivered interventions for improving mental health among young people, identify factors influencing effectiveness, and examine feasibility and acceptability. METHODS: To identify eligible interventional studies, we systematically searched 11 databas…

59Gain side

How the two sides scoreClose call

58Problem side

The gain

A meta-analysis of 11 randomized trials found chatbot-delivered interventions reduced psychological distress in people aged 10 to 24 by a small-to-moderate margin (Hedges' g = -0.28, 95% CI -0.46 to -0.10) compared with control conditions.

The problem

The same meta-analysis of seven randomized trials found chatbot interventions produced no statistically significant improvement in young people's psychological well-being (g = 0.13, 95% CI -0.16 to 0.41), and the evidence was rated low certainty.

A fixed formula sets both scores from the evidence and its sources. Reader votes don’t move them.

What the scores mean

G is the gain side, out of 100. It goes up when the gain is specific, when it measurably helped people and when the sources behind it are strong.

P is the problem side, out of 100. It goes up when the problem is specific, when it measurably hurt people and when the sources behind it are strong.

In brief

The authors searched 11 databases and search engines for studies published from January 2014 to September 2024 and found 29 interventional studies, 13 of them randomized trials. Eleven trials were usable for the meta-analysis, as two lacked sufficient outcome data. Certainty of evidence was rated moderate for distress and low for well-being.

Effects differed by design. Distress fell significantly in clinical or subclinical groups (g = -0.34) but not in nonclinical groups, and AI-based chatbots (g = -0.36) outperformed rule-based ones (g = -0.09) on distress. Only rule-based chatbots significantly improved well-being (g = 0.29). The authors describe the evidence as preliminary.

Main points

  1. Of 4,031 records screened, 29 interventional studies met the criteria; 16 were quasi-experimental and 13 were randomized trials, and 11 trials fed the meta-analysis.
  2. Distress effect: g = -0.28 across 11 trials, and leave-one-out sensitivity analyses gave estimates between -0.23 and -0.31.
  3. Well-being effect: g = 0.13 across seven trials, not statistically significant, with certainty rated low.
  4. Distress fell more in clinical or subclinical groups (g = -0.34, five trials) than in nonclinical groups (g = -0.19, six trials, not significant).
  5. Chatbots delivered through instant messengers (g = -0.29) and those using multiple communication modes (g = -0.39) cut distress; text-only chatbots did not significantly do so (g = -0.19).
  6. Attrition ranged from 0% to 89.7% (median 23.4%), and only 14 of 29 studies included safeguards such as emergency assistance or crisis identification.
  7. Most studies were run in educational settings (19) and high-income countries (18 of 29), with the US, China and New Zealand most represented.

What the review found when it combined the studies

These are the review’s own pooled results, copied from its abstract. Truvace did not recalculate them.

  • Psychological well-being in young people with chatbot-delivered interventions compared with control conditions

    Hedges' g 0.13 (95% CI -0.16 to 0.41)

    search to September 2024

    The review’s own words
    The meta-analysis indicated that chatbot-delivered interventions significantly reduced distress (Hedge's g = -0.28, 95% CI [-0.46, -0.10]), but did not have a significant effect on psychological well-being (Hedge's g = 0.13, 95% CI [-0.16, 0.41]).

  • Psychological distress with AI-based chatbots compared with control

    Hedges' g −0.36 (95% CI −0.58 to −0.14)

    8 studies · search to September 2024

    The review’s own words
    while AI‐based chatbots ( g = −0.36, 95% CI [−0.58, −0.14], n = 8) exhibited a stronger effect compared to rule‐based chatbots ( g = −0.09, 95% CI [−0.39, 0.20], n = 4)

  • Psychological distress in clinical or subclinical young people with chatbot-delivered interventions

    Hedges' g −0.34 (95% CI −0.57 to −0.10)

    5 studies · search to September 2024

    The review’s own words
    Regarding psychological distress, a statistically significant reduction was observed in the clinical/subclinical population ( g = −0.34, 95% CI [−0.57, −0.10], n = 5)

  • Psychological distress in young people with chatbot-delivered interventions compared with control conditions

    Hedges' g -0.28 (95% CI -0.46 to -0.10)

    11 studies · search to September 2024

    The review’s own words
    The meta-analysis indicated that chatbot-delivered interventions significantly reduced distress (Hedge's g = -0.28, 95% CI [-0.46, -0.10]), but did not have a significant effect on psychological well-being (Hedge's g = 0.13, 95% CI [-0.16, 0.41]).

Code checks every number above against the review’s text before it is shown. The short labels are an AI model’s reading of that text, so the review’s own sentence sits under each one. A pooled figure is only as strong as the studies behind it. How Truvace uses pooled results

The full story

The review followed PRISMA reporting and was registered with PROSPERO. Two reviewers independently screened studies and extracted data, with Cohen's kappa of 0.88 and 0.90 at the two screening stages. Risk of bias in trials was assessed with the revised Cochrane tool, and certainty with GRADE. The meta-analysis used three-level random-effects models to handle multiple effect sizes per trial.

Participants ranged from 8 to 33 years old. Twenty studies recruited nonclinical populations, five enrolled people with psychological symptoms, and four focused on young people diagnosed with physical or mental disorders. Depression and anxiety were each assessed in 14 studies. In 20 studies the chatbot was AI-based; the rest followed rules or decision trees. Cognitive behavioral therapy was used in seven studies.

On feasibility, recruitment rates ranged from 24% to 100% (median 46.9%), and 18 studies reported attrition below 30%. Participants consistently reported satisfaction, citing empathic communication, accessibility and non-judgment, but also input misunderstanding, repetitiveness and impersonal interactions. Only three studies integrated human support, and these showed lower attrition.

Heterogeneity was substantial for both outcomes (I2 of 63.94% for distress and 66.90% for well-being). The main source of bias in the trials was reliance on self-reported measures without assessor blinding. The authors recommend delivery through instant messengers, multimodal interaction, stronger privacy and security measures, longer follow-up and trials in low-income countries.

What this doesn’t fix

Most studies were in nonclinical populations in educational settings and developed countries, the trials carried risk of bias and substantial heterogeneity, and too few studies existed to assess long-term effects.

Sources

1 on file
  1. Worldviews on Evidence-Based Nursingdoi:10.1111/wvn.70059
    Peer-reviewed2025-07-15

Talk it over

Four questions to start onThis record found a gain and a problem in the same place. Which matters more to you, and who might answer differently?

Start with a question

Chatbots modestly cut distress in young people, a review of 11 trials findshttps://truvace.com/item/chatbot-delivered-interventions-for-improving-mental-health-among-young-people-a

  1. This record found a gain and a problem in the same place. Which matters more to you, and who might answer differently?

    The gain A meta-analysis of 11 randomized trials found chatbot-delivered interventions reduced psychological distress in people aged 10 to 24 by a small-to-moderate margin (Hedges' g = -0.28, 95% CI -0.46 to -0.10) compared with control conditions.

    The problem The same meta-analysis of seven randomized trials found chatbot interventions produced no statistically significant improvement in young people's psychological well-being (g = 0.13, 95% CI -0.16 to 0.41), and the evidence was rated low certainty.

    Answer in the thread

  2. How do we know? What can this kind of evidence show, and what can’t it?

    The evidence This rests on a systematic review or meta-analysis, from Worldviews on Evidence-Based Nursing.

    Answer in the thread

  3. What does it leave open? What else would you want to know before acting on it?

    The record’s own caveat Most studies were in nonclinical populations in educational settings and developed countries, the trials carried risk of bias and substantial heterogeneity, and too few studies existed to assess long-term effects.

    Answer in the thread

  4. Have you seen AI used this way where you study, work or live? Did it go the same way?

    Answer in the thread

The questions are the same on every record. The details under them are taken from this record as written.

What readers say

More from Health