Women in the UK lag men in generative-AI adoption largely because they worry more about societal risks—not because of access or skills; boosting optimism about AI could raise young women's GenAI use from about 13% to 33%, markedly narrowing the gender gap.

Main Finding

Women adopt generative AI (GenAI) substantially less than men in the UK, and this gap is driven primarily by gendered perceptions of GenAI’s societal risks (mental health, privacy, climate impact, labor-market disruption) rather than by access or capability. A composite index of those concerns explains a meaningful share of variation in adoption and is among the strongest predictors of women’s GenAI use; shifting perceptions raises young women’s use markedly and narrows the gender divide.

Key Points

• Source: nationally representative UK survey data, 2023–2024.
• Core mechanism: a composite “societal risk concerns” index (mental health, privacy, climate, labor disruption) predicts lower GenAI adoption among women.
• Quantitative importance:
  • The risk-concerns index explains between 9% and 18% of the variation in GenAI adoption.
  • It ranks among the strongest predictors of adoption for women across all ages and outperforms digital literacy and education for young women.
  • Intersectional patterns: largest gender gaps occur among younger, digitally fluent respondents with high societal-risk concerns — gender differences in personal GenAI use exceed 45 percentage points in these subgroups.
  • Counterfactual/synthetic analysis: raising optimism about AI’s societal impact increases GenAI use among young women from 13% to 33%, substantially narrowing the gender gap.
• Interpretation: differences in social/ethical perceptions (risk aversion, trust, value judgments) are the primary drivers of unequal GenAI uptake, not differential access or digital skills.

Data & Methods

• Data: nationally representative UK survey waves collected in 2023–2024 that measure GenAI awareness and personal use, demographic and socioeconomic covariates, digital skills, and attitudes about AI’s societal impacts.
• Key variable: composite index of societal risk concerns combining items on mental-health risks, privacy concerns, environmental/climate impact, and fears about labor-market disruption.
• Empirical approach:
  • Multivariate regression analyses to assess predictors of GenAI adoption, controlling for demographics, education, digital literacy, occupation, and access.
  • Predictor-ranking to compare the explanatory power of the risk index versus other covariates (e.g., digital literacy, education).
  • Intersectional subgroup analyses by age and digital fluency to reveal heterogeneous gender gaps.
  • Synthetic twin panel design: construct matched counterfactual “twins” (synthetic panel) to simulate how changes in AI-related optimism would affect adoption — used to estimate increase in young women’s use from 13% to 33% under increased optimism.
• Robustness: results hold across age groups and after accounting for access and capability measures (suggesting perceptions, not constraints, drive the gap).

Implications for AI Economics

• Adoption and productivity: gendered perception gaps may produce unequal adoption of productivity-enhancing GenAI tools, potentially reducing aggregate productivity gains and creating gendered productivity differentials within firms and occupations.
• Skills and human capital formation: lower early adoption among women — especially young, digitally fluent women — could alter skill accumulation, career trajectories, and comparative advantage as AI reshapes tasks and required skills.
• Labor-market inequality: differential uptake driven by attitudes (not access) risks widening existing gender wage and employment gaps as AI complements some tasks and skills.
• Policy and firm intervention levers:
  • Address perceptions directly (communication, education, demonstrations) rather than focusing solely on access or digital skills training.
  • Design outreach and interventions targeted at subgroups with high societal-risk concerns (young women, digitally fluent) to reduce misperceptions and build trust.
  • Improve governance, transparency, and safety assurances (privacy protections, ethical AI practices) to lower perceived societal risks and encourage broader, more inclusive adoption.
• Research and evaluation: interventions that change beliefs about AI’s societal impacts (information treatments, institutional safeguards, product design) merit experimental testing to assess causal effects on adoption and downstream economic outcomes.