Truace
Subscribe
55/100

Moderate evidence

Cite
More
Export CSVExport JSONEmailLinkedInX

Survey of 19,484 Chinese students links problematic AI use to depression, anxiety, poor sleep

Study title: Initial Psychometric Evaluation of the Problematic Use of Generative Artificial Intelligence Scale Among Chinese School Students

Evidence type
Survey
Sample
n = 19,484
Sources
1 sourcePeer-reviewed
Published
2 min read
Truvace record
TRV-2026-2675cite or check it

Study title: Initial Psychometric Evaluation of the Problematic Use of Generative Artificial Intelligence Scale Among Chinese School Students

Researchers surveyed 19,484 students in Grades 4–9 across 72 schools in Wenzhou, China, who had used generative AI, to test a short scale for problematic use. The final 8-item scale scored higher in students with more depressive symptoms (r = 0.396), anxiety (r = 0.387), insomnia (r = 0.365) and short-video addiction (r = 0.413).

Read the study abstract

Generative artificial intelligence (GAI) is increasingly used by school students, yet validated instruments for assessing problematic GAI use (PUGAI) in this population remain scarce. This study evaluated the psychometric properties of the Problematic Use of Generative Artificial Intelligence Scale (PUGAIS) in a large school-based sample of Chinese students. A cross-sectional survey included 19,484 students in Grades 4–9 who had used GAI. The sample was randomly split for exploratory and confirmatory factor anal…

The problem

In a survey of 19,484 Chinese students in Grades 4–9 who had used generative AI, higher problematic-use scores were linked to more depressive symptoms (r = 0.396), anxiety (r = 0.387) and insomnia (r = 0.365), and 7.2% fell in a high-score group.

In brief

A latent class analysis put 7.2% of students in a high-score group, with 55.2% in a low-score group and 37.5% in a moderate group. Nearly half of the 40,120 returned questionnaires (47.77%) came from students who had never used generative AI and were excluded.

The findings are correlations from one time point, so they cannot show that AI use caused distress or poor sleep. The scale had a floor effect, with 37.8% of students scoring the minimum, and the proposed cutoff of 20.5 was derived from the same responses and is not a diagnostic threshold.

Main points

  1. The survey ran in March 2025 across 72 schools (51 primary, 21 secondary) in Wenzhou, with students averaging 12.14 years old.
  2. The scale was adapted from the 10-item Smartphone Addiction Scale—Short Version; two items were dropped to give an 8-item version.
  3. Internal consistency was high (Cronbach's α = 0.940), and item thresholds and loadings held across sex, school type and age groups.
  4. Fit evidence was mixed: RMSEA was 0.142 in the validation subsample, so a single-factor structure is only partly supported.
  5. 37.8% of students (7,363) scored the minimum, which limits the scale's ability to separate students at the low end.
  6. The 7.2% high-score group and the 20.5 cutoff came from the same item responses, so they are descriptive rather than diagnostic.
  7. The authors say frequent or long AI use for schoolwork should not count as problematic on its own.

The full story

The team adapted the 10-item Smartphone Addiction Scale—Short Version to generative AI, with translation and review by behavioural scientists, psychologists, a clinical psychologist and a psychiatrist. No formal content-validity check, cognitive interviewing or pilot test was done. Items used a 4-point agree–disagree rating.

Of 40,120 questionnaires returned, 19,165 were excluded because students had never used generative AI, and a further 1,471 for missing data, leaving 19,484 (52.8% boys, 49.5% primary school). The sample was split into 9,761 for exploratory analysis and 9,723 for validation. The original 10-item model fit poorly, and items 1 and 2 were removed, leaving an 8-item scale.

Sensitivity analyses showed the removal of items 1 and 2 depended on the statistical treatment: with ordinal correlations, those items loaded higher. Even so, the 8-item version fit better than the 10-item version (CFI 0.983 versus 0.967). Items 5 and 6 correlated at 0.800, which may mean overlapping content.

Latent class analysis kept a three-class solution, replicated in both subsamples. The ROC analysis gave an AUC of 0.998 and a cutoff of 20.5, but the authors say this only reflects internal correspondence, and that any use for screening or assigning students to support needs independent validation.

Sources

1 on file
  1. Behavioral Sciencesdoi:10.3390/bs16101763
    Peer-reviewed2026-09-28

Talk it over

Four questions to start onWho carries this cost, and what would have to change to prevent it?

Start with a question

Survey of 19,484 Chinese students links problematic AI use to depression, anxiety, poor sleephttps://truvace.com/item/initial-psychometric-evaluation-of-the-problematic-use-of-generative-artificial-

  1. Who carries this cost, and what would have to change to prevent it?

    The problem In a survey of 19,484 Chinese students in Grades 4–9 who had used generative AI, higher problematic-use scores were linked to more depressive symptoms (r = 0.396), anxiety (r = 0.387) and insomnia (r = 0.365), and 7.2% fell in a high-score group.

    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 survey with a sample of 19,484, from Behavioral Sciences.

    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 The study was cross-sectional, drawn from one city in Zhejiang Province, and had no independent clinical criterion, so the high-score group and 20.5 cutoff are unvalidated and causation cannot be inferred.

    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 Education