Research: Women Less Likely to Use Generative AI Over Ethical Worries

Oxford University research reveals that women are significantly less likely to use generative AI than men. The reason isn't a lack of skill, but greater concern about AI's potential negative impacts on employment, privacy, mental well-being, and society as a whole.

As the primary subjects of unauthorized deepfake content, women have been at the forefront of activism against this controversial application of generative AI for the past seven years, achieving several notable recent victories.

However, a new Oxford University study suggests this portrayal of women's AI concerns is too limited. It finds women's use of all forms of generative AI lags behind men's, not due to access or skill gaps, but because they are more likely to perceive AI as harmful to mental health, jobs, privacy, and the environment.

The paper states:

‘Using nationally representative UK survey data from [2023–2024], we demonstrate that women adopt GenAI substantially less often than men due to differing perceptions of its societal risks.

‘Our composite index measuring concerns about mental health, privacy, climate impact, and labor-market disruption explains 9-18% of adoption variation and ranks among the strongest predictors for women across all age groups–surpassing digital literacy and education for young women.’

The most significant gaps, researchers note, occur among younger, tech-savvy users who express strong concern about AI's social risks, with gender differences in personal use exceeding 45 percentage points:

The gender gap in frequent generative AI use is largest among women with high digital skills who also express strong concerns about mental health, climate, privacy, and job market risks. The smallest gaps are found among those more optimistic about AI's societal effects. Source

By comparing similar respondents across successive survey waves in a synthetic-twin panel, the study finds that when young women become more optimistic about AI's societal impact, their generative AI usage increases from 13% to 33%, substantially narrowing the gap. Among those concerned about climate harm, the gender gap in generative AI use widens to 9.3 percentage points, and among those worried about mental health harm, it expands to 16.8 points. This change is driven not by increased male usage, but by significant declines among women.

The authors identify a distinct cultural effect related to gender*:

‘On average, women exhibit greater social compassion, traditional moral concerns, and a pursuit of [equity]. Meanwhile, moral and social concerns have been found to influence technology acceptance.

‘Emerging research on GenAI in education indicates women are more likely to view AI use in coursework or assignments as unethical, equivalent to cheating, facilitating plagiarism, or spreading misinformation.

‘Greater concern for social good may partly explain women’s lower adoption of GenAI.’

The authors suggest the women's perspective observed in the study is valid:

‘[Women’s] heightened sensitivity to environmental, social, and ethical impacts is justified: generative AI systems currently have significant energy demands, uneven labor practices, and well-documented risks of bias and misinformation.

‘This suggests that narrowing the gender gap requires not only shifting perceptions but also improving the underlying technologies. Policies that incentivize lower-carbon model development, strengthen safeguards against bias and wellbeing harms, and increase transparency around supply-chain and training-data practices would address legitimate concerns—while ensuring women’s risk awareness drives technological improvement rather than acting as an adoption barrier.’

They further note that while the study clearly demonstrates this adoption gap, its findings are likely even more pronounced outside the UK, where the research was conducted.

The new paper, titled ‘Women Worry, Men Adopt: How Gendered Perceptions Shape the Use of Generative AI’, features researchers from the Oxford Internet Institute, the Institute for New Economic Thinking in Belgium, and the Humboldt Institute for Internet and Society in Berlin.

Data and Approach

Recent research trends indicate women are using all forms of generative AI less frequently than men, despite equal ability and access. This shortfall is estimated to contribute to the gender wage gap, consistent with prior trends linking lower internet use among women to lower salaries:

From the 2023 paper ‘Has Internet Usage Really Narrowed the Gender Wage Gap?: Evidence from Chinese General Social Survey Data’, an illustration of internet use narrowing the gender wage gap more significantly at lower wage levels, with diminishing returns as wage levels rise. Source

For the new study, the authors used year-on-year data from the UK government's Public attitudes to data and AI: Tracker survey to analyze how perceptions of AI-related risks influence adoption patterns across genders, identifying risk sensitivity as a key factor in women's reduced usage.

GenAI gender gaps widen significantly when risk concerns combine with other traits. The largest gap, illustrated below, is 5.3 points, occurring among women with high digital skills who perceive AI as a mental-health risk:

Gender gaps in GenAI use vary based on attitudes and demographics. Red cells indicate where men use GenAI more than women, particularly for personal use. The largest gaps occur when high digital skills combine with mental health risk concerns. In work settings, gaps widen with privacy or climate concerns. Blue cells denote smaller or reversed gaps.

Mental‑health concerns tend to amplify the gender gap across most groups, with the strongest effect among younger, tech-savvy users. Privacy worries also widen the divide, pushing the gap as high as 22.6 points in some work contexts.

Even among older respondents concerned about AI's climate impact, the gap remains substantial at 17.9 points, indicating perceptions of harm weigh more heavily on women—including in groups where overall AI use is relatively low.

Risk Perceptions

To determine how strongly risk perception influences adoption, researchers created a composite index based on concerns about AI's effects on mental health, climate, privacy, and employment. This score was tested alongside education, occupation, and digital literacy using random forest models split by age and gender. They found that across all life stages, AI-related risk perceptions consistently predicted generative AI use—often ranking higher than skills or education, especially for women:

Random forest models, stratified by age and gender, reveal AI-related risk perception is a stronger predictor of generative AI use for women than men, ranking among the top two factors across all female age groups and exceeding the influence of digital literacy and education. For men, digital literacy is dominant, while risk perception ranks lower and is less consistent. The models show societal concerns shape AI adoption far more strongly for women than traditional skill or demographic factors. Please refer to the source PDF for better legibility and resolution.

Across all age groups, concern about AI's societal risks predicted generative AI use more strongly for women than men. For women under 35, risk perception was the second most influential factor shaping use, compared to sixth for men. Among middle-aged and older groups, it ranked first for women and second for men.

Across models, risk perception accounted for 9% to 18% of predictive importance, outweighing education and digital skill measures.

According to the paper, these results indicate women's lower generative AI adoption stems less from personal risk concerns and more from broader ethical and societal considerations. The hesitation appears driven by greater awareness of AI's potential harm to others or society, rather than to themselves.

Synthetic Twins

To test whether changing attitudes on these topics can shift behavior, researchers used a synthetic-twin design, pairing similar respondents across two survey waves. Each person from the earlier wave was matched with a later respondent of the same age, gender, education, and occupation.

The team then compared changes in generative AI use among those who either improved digital skills or became more optimistic about AI's societal effects. This allowed them to isolate whether greater literacy or reduced concern could increase adoption, especially among younger adults:

To test whether targeted changes affect AI use, researchers compared young adults who improved digital skills or became more optimistic about AI’s societal impact. Both changes increased adoption, but digital literacy widened the gender gap by benefiting men more. In contrast, greater optimism boosted women’s usage from 13% to 33%, narrowing the divide and suggesting addressing ethical concerns may be more effective than skill-building alone.

Boosting digital literacy increased generative AI use for both genders but widened the gap, with men benefiting more. In the full sample, women's usage rose from 9% to 29%, while men increased from 11% to 36%.

Among younger adults, gains in digital literacy raised men's usage sharply from 19% to 43%, while women's increase from 17% to 29% was modest and not statistically significant. Conversely, greater optimism about AI's societal impact produced a more balanced shift, with women rising from 13% to 33% and men from 21% to 35%. In the full sample, women moved from 8% to 20%, and men from 12% to 25%.

Therefore, while digital upskilling boosts overall adoption, it also tends to widen gender gaps. Reframing perceptions of AI's broader impact appears more effective at increasing women's use without disproportionately boosting uptake among men.

Conclusion

The significance of these findings evolves as the paper progresses. Initially, as quoted above, the authors commend women's greater global concern and ethical stance. Later, a more pragmatic viewpoint emerges, questioning whether women might be 'left behind' due to their moral vigilance and reservations:

‘[Our] findings point to broader institutional and labour-market dynamics. If men adopt AI at disproportionately higher rates during the period when norms, expectations, and competencies are still taking shape, these early advantages may compound over time, influencing productivity, skill development, and career progression.’

* My conversion of the authors’ inline citations to hyperlinks.

First published Thursday, January 8, 2026