Skills, Mindsets, and AI

This study examines whether brief, scalable online interventions can reduce anxiety and resistance toward artificial intelligence (AI) in the workplace and increase openness to reskilling. Many workers report feeling stressed or threatened by AI, even though existing evidence suggests limited negative effects on overall employment or mental health. Such perceptions can slow adoption of AI tools and reduce engagement with reskilling opportunities.

The present study constitutes the second wave of a longitudinal research design. In a prior study, “Making Workers AI Ready: Behavioral Interventions and AI Adoption Among Workers” (AEARCTR-0018144), workers aged 18–65 completed a baseline survey along with an incentivized, occupation-specific writing task within a 2×2 experimental framework that varied both an AI-use permission statement and an informational message emphasizing AI’s productivity benefits. The current study re-contacts participants from the United States who took part in the earlier study and invites them to complete a follow-up survey approximately one month after the initial intervention.

In this study, participants are randomly allocated to one on the four conditions corresponding to two interventions and their respective control conditions. We blocked randomization by treatment conditions in the prior study, so that control and treatments are balanced within each prior study conditions.

Participants assigned to the synergistic mindset intervention complete a brief online module adapted from Yeager et al. (2022). The module combines two evidence-based components: a growth mindset, defined as the belief that abilities and skills can be developed with effort, practice, and learning (Dweck, 2006), and a stress-is-enhancing mindset, defined as the idea that the body's stress response can be reframed as a source of energy and focus that supports performance and growth. The intervention consists of a short video, interactive content, and reflective writing exercises that help participants apply these ideas to challenges related to AI and workplace change, emphasizing that adapting to new technologies such as AI can be stressful but also an opportunity for learning and skill development.

Participants assigned to the reskilling intervention complete a short, structured guide to effective prompting. This video introduces practical strategies for interacting productively with a generative AI tool, including how to formulate clear prompts, provide relevant context, and iteratively refine outputs. The goal of the reskilling intervention is to lower barriers to AI use, increase self-efficacy, and encourage hands-on experimentation with AI tools during simple productivity tasks.

Participants assigned to the mindset-control condition complete a module of similar length, tone, and engagement presenting neutral neuroscience content about brain anatomy and basic functions, without motivational or learning-related framing. Participants assigned to the reskilling control condition view a neutral educational video on the history of AI prior to the 21st century, which provides general background information but does not involve skill acquisition or guidance on AI use.

All modules are designed to be completed online in approximately 7 minutes. After completing the assigned module, participants again perform short productivity tasks, giving all the participants the possibility to use a generative AI tool. Outcomes measured include AI-related attitudes (measured using the ATTARI-12 scale), trust in AI, self-efficacy, AI use during tasks (subjective and objective metrics), and task performance.

By combining a psychological mindset framing with a prompt-based reskilling, each evaluated relative to appropriate control conditions, this study provides early experimental evidence on whether brief interventions can increase acceptance of AI and support healthier, more proactive adaptation to technological change in the workplace.

Sample and Implementation
Ideally, this study aims to collect up to 200/250 observations per experimental treatment group. Participants will be allocated across the four experimental groups, two treatments (targeting approximately 200/250 per arm) and two neutral sub-control arms (targeting approximately 200/250 per arm). Pending funding, the sample will be expanded to increase statistical power and provide more robust estimates of intervention effects.

2026-05-19

2027-05-31

The study will measure how a brief synergistic-mindset and reskilling intervention affects workers' attitudes, confidence, and performance related to artificial intelligence (AI).

1) Task performance:
Performance (productivity) in short writing or coding tasks, measured by accuracy, quality ratings, and completion time.

2) Behavioral outcomes:
AI adoption and use. Willingness to engage with AI tools, measured by whether participants choose to use a generative AI tool during productivity tasks, as well objective measure of AI use.

3) Attitudinal outcomes:
General attitudes toward AI, measured using the ATTARI-12 scale (cognitive, affective, and behavioral subcomponents).
Trust and perceived reliability of AI tools (e.g., "I can rely on AI to improve my work").
Self-efficacy for learning and using new digital tools (12-item SECS scale).
Growth mindset items (e.g., "I can get better at using AI through practice and effort").

4) Learning and reskilling intentions:
Intentions to invest in AI-related learning or training, measured by stated money willingness to spend on improving growth mindset and AI skills.

Outcomes are measured in order to assess the immediate impact of the mindset and reskilling interventions.

We will adjust for multiple hypothesis testing by organizing outcomes into pre-specified families and controlling the false discovery rate (FDR) within each family using the Benjamini-Hochberg procedure (Benjamini and Hochberg, 1995; Anderson, 2008).

The study will also explore a set of secondary and exploratory outcomes to better understand the mechanisms through which the mindset and reskilling interventions may influence adaptation to AI-driven change.

Psychological mechanisms and beliefs: Feelings of empowerment, enjoyment, difficulty, guilt, cheating, imposter syndrome, sense of agency, loss aversion, ethical/privacy/transparency concerns when working with or alongside AI. Beliefs about AI's labor market impacts on workers by age and gender and working experiences (displacement, working hours/days, collaboration at work).

Behavioral engagement: Engagement and persistence in completing AI-related tasks (e.g., time spent, number of attempts, dropout). Open-ended reflections coded for mindset-consistent language (e.g., references to learning, effort, or reappraising stress). Self-reported AI use outside the study (e.g., frequency of AI-assisted work or learning activities during the follow-up period).

Heterogeneity: Differences in effects on the outcomes by occupation, education, age, gender, socioeconomic characteristics, baseline AI attitudes (i.e., feelings of guilt, ethical/privacy/transparency concerns, ATTARI, self-efficacy), prior AI exposure/familiarity (Humlum and Vestergaard, 2025), loss aversion, risk preferences, trust, time preferences, Big Five personality traits, cognitive skills, collaboration at work, and planned retention with current employer.

Life satisfaction, wellbeing, physical and mental health.

This study uses a randomized online experiment to test whether brief psychological and educational interventions can improve workers' performance, attitudes and behaviors toward artificial intelligence (AI).

Longitudinal Study Design
The study follows a longitudinal, two-survey design. In the prior study, participants completed baseline measures and were randomly assigned to one of four conditions :1) receive either a brief informational nudge highlighting evidence that AI use can increase productivitye; 2) receive explicit permission to use AI; 3) receive both the informational nudge and explicit permission; 4) a control condition with no information and no permission. After this manipulation, participants completed a short productivity task.

In this study we are re-contacting participants and cross-randomizing them across four groups, corresponding to the two interventions (i.e., growth mindset and reskilling) and their respective control conditions. Randomization occurs at the individual level within occupation and stratified by the previous study treatment arms.

1) Synergistic Mindset Intervention (Treatment 1):
Participants assigned to this condition complete a brief online synergistic mindset video adapted from Yeager et al. (2022). The video combines two scientific ideas: a growth mindset, defined as the belief that abilities can improve through effort and learning; and a stress-is-enhancing mindset, defined as the belief that stress can be reframed as energy that supports focus and performance. Participants complete short readings and afterwards complete a reflective writing exercise applying these ideas to challenges related to AI and workplace change. The duration of the video is approximately 7 minutes.

2) Neuroscience Video (Control 1):
Participants in this condition complete a module of similar length, tone, and engagement containing neutral neuroscience content, such as facts about brain anatomy and function, without motivational framing or reference to AI.

3) Reskilling Intervention (Treatment 2):
Participants assigned to this condition watch a short video: a structured guide to effective prompting, which provides practical guidance on how to interact productively with a generative AI tool, including how to formulate clear prompts, provide relevant context, and iteratively refine outputs. The duration of the video is approximately 7 minutes.

4) History of AI prior to the 21st Century Video (Control 2):
Participants in this condition view a neutral educational video on the history of AI prior to the 21st century, which provides general background information but does not involve skill acquisition or guidance on AI use.

Outcome Measures:
After completing the assigned module, all participants perform an occupation-specific incentivized productivity task, where they complete a writing or coding tasks with optional use of an AI assistant. We then measure outcomes as described above (primary and secondary outcomes).

Sample and Implementation
Ideally, the study aims to collect up to 200/250 observations per experimental treatment group.

Not available

Participants will be individually randomized using the randomization functions embedded in the online survey platform (e.g., Qualtrics or Prolific assignment tools). Randomization will be implemented automatically at the time participants begin each survey to ensure allocation concealment.

Participants will be randomly allocated across the four experimental groups, two treatments (reskilling and mindset interventions) and two neutral sub-control arms (neutral history of AI and neuroscience content).

Randomization in this study will be stratified by each cell of the 2x2 design based on the previous study titled “Making Workers AI Ready: Behavioral Interventions and AI Adoption Among Workers” (AEARCTR-0018144).

The unit of randomization is the individual participant.

In this study, each participant is independently assigned to one of four conditions: Mindset Treatment, Mindset Control, Reskilling Treatment, or Reskilling Control. Randomization in this study is stratified by each cell in the previous study.

No

N/A

Synergistic Mindset Intervention: ~150-250 participants Reskilling Intervention: ~150-250 participants Neuroscience Control: ~150-250 participants History of AI before the 21st Century Control: ~150-250 participants Total: ~600 to 1000 participants

Synergistic Mindset Intervention: ~150-250 participants

Reskilling Intervention: ~150-250 participants

Neuroscience Control: ~150-250 participants

History of AI before the 21st Century Control: ~150-250 participants

Total: ~600 to 1000 participants