Does AI literacy keep people engaged at work?
A peer-reviewed Chinese study followed 1,108 employees for a year and found that higher baseline AI literacy predicted later work engagement. The design strengthens the timing evidence, but it does not show that AI training caused the difference.
Editorial responsibility: The Impact of AI Editorial Desk · Report a factual concern
Research topic
Whether employee AI literacy predicts work engagement over one year, and the possible roles of self-efficacy and burnout
At a glance
- 1The study retained 1,108 of 1,212 employees across three survey waves over one year, with participants drawn from internet companies, pharmaceutical firms, publishing groups and high schools in China.
- 2Higher baseline AI-literacy scores predicted work engagement a year later after adjustment for baseline engagement, age, gender and reported AI-use intensity; the direct standardized effect was 0.12 after the mediators were included.
- 3This is longitudinal observational evidence, not a training trial: self-reported literacy may capture other advantages, and the study cannot establish that improving AI literacy will cause engagement to rise.
Living evidence record
Impact record IAI-060GS9K
Evidence stage
Studied
Confidence
Supported
Reporting basis
Source analysis
Independent support
Present
Record status
Monitoring
Last checked
2 October 2026
Source trail
2 direct sources across 1 source type.
People impact
Documented in this record.
Uncertainty
Limits and next checks are explicit.
Stages describe the evidence available—not whether a technology is good or bad. See the public method.
Related-source reporting disclosure
This record analyses 2 linked source records around the same underlying development. The extra records add method, date or context, but they do not by themselves constitute independent replication of every performance claim or predicted outcome.
The study separates literacy from simple tool use
Workplace debates often treat AI adoption as a yes-or-no question: an employee either uses a tool or does not. The study by Qingqi Liu, Yuju Lei and Jingjing Li asks a more useful question. Does a broader ability to recognise AI, use it, judge its limitations and apply it ethically predict whether people remain energetic, dedicated and absorbed in their work? That distinction matters because frequent use can coexist with weak understanding, misplaced confidence or exhaustion.
The authors define AI literacy through four self-reported dimensions: awareness, usage, evaluation and ethics. Their outcome is work engagement, measured with the nine-item Utrecht Work Engagement Scale. They also test two possible psychological pathways six months into the study: work self-efficacy, meaning confidence in handling work problems, and burnout, covering exhaustion, depersonalisation and reduced accomplishment. This design does not measure productivity, pay, promotion or job retention.[1][2]
Three waves retained 1,108 employees over a year
At the first wave in June 2024, 1,212 employees from internet companies, pharmaceutical firms, publishing groups and high schools completed the survey. Six months later, 1,156 returned; by June 2025, 1,108 remained for the formal analysis. The final group included 618 men and 490 women aged 21 to 60, with a mean age of 40.63 years. The paper does not provide a province-by-province or employer-by-employer denominator, limiting how precisely readers can judge occupational representation.
Reported AI exposure varied substantially. A quarter of participants said they used AI products for less than ten minutes a day, 36.7% for ten to 30 minutes and 18.2% for 31 to 60 minutes. Another 12.1% reported one to two hours, while 7.8% reported more than two hours. The researchers compared people who remained with those who dropped out on the core measures available at each stage and found no statistically significant differences, although such checks cannot rule out every form of attrition bias.[2]
The timing is stronger than a one-off workplace survey
Baseline AI literacy and work engagement were measured at Time 1. Work self-efficacy and burnout were measured at Time 2, six months later. Engagement was measured again at Time 3, one year after baseline. The analysis adjusted for gender, age, AI-use intensity and initial work engagement. That last adjustment is important: it asks whether literacy predicts change beyond the fact that some people were already more engaged at the beginning.
The researchers used a serial mediation model with 5,000 bootstrap samples. Continuous variables were standardised, so the reported coefficients describe changes in standard-deviation units rather than minutes worked or points of productivity. The three-wave sequence establishes temporal order for the predictor and later outcome. It does not, however, create random assignment, eliminate unmeasured confounding or prove that self-efficacy precedes burnout because both mediators were recorded at the same six-month wave.[2]
Higher literacy predicted later engagement, with a modest direct effect
Before the mediators were added, baseline AI literacy predicted engagement one year later with a standardised total effect of 0.31 and a 95% confidence interval from 0.26 to 0.36. In the full model, the remaining direct effect was 0.12, with a 95% confidence interval from 0.07 to 0.17. The result is statistically clear within this sample, but it should not be translated into a claim that an AI course raises engagement by 12% or 31%; those figures are standardised model coefficients, not percentage changes.
AI literacy also predicted greater work self-efficacy at six months (0.36) and lower burnout (-0.18). Self-efficacy predicted later engagement positively (0.35), while burnout predicted it negatively (-0.23). The largest indirect pathway ran through self-efficacy alone, estimated at 0.13. Burnout alone contributed an indirect effect of 0.04, and the proposed self-efficacy-to-burnout chain contributed 0.02. Confidence intervals excluded zero, but mediation remains a model of the observed associations rather than proof of the underlying mechanism.[2]
Self-reporting is both the study's instrument and its main constraint
AI literacy was measured with a 12-item scale validated for Chinese people aged 17 to 65. Participants rated statements about identifying AI, using it to improve efficiency, evaluating capabilities and limitations, and following ethical principles. Internal consistency was acceptable at 0.78. Self-efficacy, burnout and engagement measures reported higher reliability. Reliable questionnaires can distinguish response patterns, but they do not demonstrate that a person can detect hallucinations, protect confidential data or complete a job task with AI.
All major variables came from the same participants. People who see themselves as capable and engaged may also rate their AI literacy more positively. Employer support, job autonomy, education, income, occupation, access to better tools and managerial culture could influence both literacy and engagement. The model adjusts for a limited set of covariates, not all of these factors. The study therefore supports prediction over time, not a causal estimate of what an intervention would achieve.[2]
The practical message is to test training, not merely mandate it
For employers, the finding makes a reasonable case for treating AI literacy as more than prompt tips. Training should include tool boundaries, verification, data handling and ethical use, alongside guided practice on real tasks. A programme can be evaluated with pre-specified outcomes: demonstrated skill, error detection, confidence calibrated to actual performance, workload, burnout and engagement. Managers should compare the results with a suitable control group instead of assuming that attendance equals capability.
For employees, literacy may reduce uncertainty by making it easier to decide when AI helps, when a human check is necessary and when a tool should not receive sensitive information. Yet the paper does not justify shifting responsibility entirely onto workers. Poorly chosen systems, surveillance, unrealistic output targets and inadequate staffing cannot be repaired by individual training. Organisational safeguards and job design remain part of whether AI is experienced as a useful resource or an additional demand.[1][2]
The evidence comes from one national and cultural setting
The authors explicitly caution that the results were obtained in China and that the literacy scale was validated for Chinese young and middle-aged adults. Workplace cultures, labour relations, regulatory expectations and organisational support differ across countries and sectors. The sample includes four broad employer types, but the paper does not show that it represents the Chinese workforce or that the effects are similar for frontline, professional, managerial and precarious workers.
The work was funded by Chinese national, provincial and education research programmes. The authors state that they have no competing interests, and the study received ethics approval from Beijing Normal University. Funding by public research bodies does not remove design limitations, but the disclosed interests do not indicate a commercial vendor evaluating its own training product. The accepted article is peer reviewed and citable while still awaiting the publisher's final copy-editing.[2]
What would change the assessment
The strongest next test would randomly assign employees or teams to a defined AI-literacy programme, an attention-control programme and usual practice. Researchers should measure demonstrated skills rather than self-ratings alone, record tool access and job design, and follow engagement, burnout, performance and retention for at least a year. Cluster assignment could reduce contamination when colleagues share training. Independent replication should include countries with different labour institutions and workplaces with limited digital infrastructure.
This study moves the evidence beyond a one-time correlation: literacy came first, and engagement was measured again a year later. That warrants attention, especially because self-efficacy accounted for the largest modelled pathway. The assessment would become causal only if well-controlled interventions reproduce the effect, show that competence actually improves and rule out the possibility that already advantaged or supported employees simply report both higher literacy and stronger engagement.[2]
What this means for people
- Workers may benefit when training turns uncertain tool use into calibrated, practical competence.
- Employers could waste time or shift risk onto staff if they treat self-rated literacy as proof of safe performance.
- Unequal access to training, protected practice time and approved tools could widen existing workplace advantages.
Global context
The study was conducted in China across four broad employer groups. Its longitudinal design is valuable, but labour institutions, tool access and workplace cultures differ internationally. The result should be treated as a strong Chinese predictive association that requires experimental and cross-country replication, not a universal training effect.
What the evidence does not yet show
- The observational design cannot establish that AI literacy causes higher engagement.
- AI literacy and all psychological outcomes were self-reported rather than demonstrated in performance tasks.
- Both proposed mediators were measured at the same wave, so their short-term order cannot be established.
- The Chinese sample and incomplete occupational breakdown limit generalisation to other labour markets and job types.
What to watch next
- Randomised workplace trials that test defined AI-literacy programmes against suitable controls.
- Objective measures of verification skill, safe data handling and task performance alongside engagement.
- Replications across occupations, countries, contract types and levels of worker autonomy.
- Whether training benefits persist without increasing workload, monitoring or inequity in access to tools.
Evidence trail
Sources used for this report
Links checked 2 October 2026
This report is labelled source analysis. We summarise and analyse source material in our own words; company statements remain attributed claims until independently supported. Translated summaries preserve the meaning of the original source and link back to it. Read our editorial standards.
Continue the story
Related reporting
Work & Skills
ILO says safe AI use is becoming a basic workplace skill
The International Labour Organization's global skills report argues that technical AI jobs remain a small niche while a much wider workforce increasingly needs the ability to understand and use AI safely and ethically.
4 min · 1 source
Work & Skills
Who can claim the UK's free AI and technology training—and what does it prove?
The Government Digital Service has opened more than 200 free learning and certification pathways to eligible UK civil and public servants until 31 December. Last year's programme recorded over 10,000 registrations and 1,200 certifications, but the government has not published completion, skill-gain or workplace-outcome data.
6 min · 2 sources
Work & Skills
If robots can do much physical work, why are so few tasks economical to automate?
Anthropic researchers rate 7,594 physical job tasks and estimate that robots can perform 74% in some setting, yet are currently cheaper than people for only 0.3% of all work. The gap between technical exposure and economic adoption is the result that matters for workers.
6 min · 1 source
Reader discussion
Add evidence, experience or a question
No account is required. Reader notes are published after a brief civility, relevance and safety check; disagreement is welcome.
Published reader notes
0No published reader notes yet. You can start the evidence-led discussion above.
Prefer a private correction or response? Contact the newsroom.