A sweeping review finds persistent gender gaps in generative AI adoption across countries, jobs, and tools.
Listen to this article:
This Insight Article revises and updates the version published on April 17, 2025.
The rapid spread of generative AI has created enormous expectations for productivity and growth. Yet new research points to a troubling pattern: women are less likely than men to adopt and intensively use these tools. In a new version of the HBS working paper “Global Evidence on Gender Gaps and Generative AI Over Time,” a team of co-authors including HBS AI Institute PI Rembrand Koning document that women continue to adopt and use generative AI at lower rates than men across countries, occupations, and tools. Drawing on a larger body of evidence than the 2024 version of their paper, the updated study shows that while the gender gap has narrowed as generative AI has diffused, it has not disappeared. This raises urgent questions about whether this technology will reduce inequality or quietly reinforce it.
Key Insight: A Near-Universal Divide in AI Adoption
“Our results reveal a remarkably consistent gap.” [1]
The authors’ first contribution is scope. They assembled a systematic review of 76 sources, including academic studies, industry surveys, and government reports, spanning more than 100 countries and covering over 300,000 people for whom both gender and AI usage were known. The takeaway result is a sample-size-weighted AI adoption rate of 47.8% for men versus 39.3% for women. In relative terms, between 2023 and 2026, men have been approximately 22% more likely to report using generative AI. The authors find that the gap has narrowed over time, especially since the earliest period after ChatGPT’s launch, but it has not disappeared. Since early 2025, the relative gap appears to have stabilized around 16 percent.
Key Insight: The Gap Persists
“To further unpack this pattern, we turn to web analytics data that allow us to track trends by month, country, and tool, providing finer grained analyses and allowing us to address potential concerns around publication bias.” [2]
One possible explanation is that men and women work in different occupations, and men are overrepresented in fields where AI is more visible or useful. However, gender differences appear even within the same occupations, organizations, and educational settings. For instance, among full-time software engineers at a single global technology company, 43% of male engineers used an internal AI coding tool at least once compared to just 31% of female engineers. Among Danish workers in 11 AI-exposed occupations drawn from a nationally representative sample, women were 16 percentage points less likely than men in the same occupation to have used ChatGPT for work.
The gap also appears in usage intensity. Women spend less time per visit on six of the top ten AI tools in the U.S. panel. On ChatGPT, women’s visits are 8.9% shorter; on Gemini, 10.7% shorter; and on Grok, 15.9% shorter. Observational data reveals less intensive AI use among women across different settings: compared to their male counterparts, female students write shorter prompts and give up faster when AI fails, while female software engineers submit fewer prompts and copy less generated code. However, this pattern is not universal, as women actually spend more time than men on conversational or companion-oriented tools, though web traffic to these tools is still male-skewed.
Key Insight: Mechanisms of Inequality
“These mechanisms differ in their implications as AI use becomes more mainstream and continues to diffuse.” [3]
The researchers divide the underlying causes of the gap into five mechanisms (or frictions). The first is knowledge and familiarity. Women consistently report lower self-assessed knowledge of generative AI, score lower on objective LLM knowledge tests, and more often cite not knowing how to use AI as a barrier. The second friction is perceived usefulness. Women are less likely to believe AI will benefit their careers or boost their productivity, and more likely to be uncertain about the returns. The third friction is institutional support, training, and confidence. Women report weaker employer encouragement, less workplace training, and lower confidence in prompting AI effectively. They are also more likely to report needing training before they can benefit from AI. A UK pilot found that short hands-on training substantially increased women’s usage, which is encouraging, but also reveals how much the default environment is working against them. The fourth friction is social legitimacy and status costs. Women are more likely to worry that using AI will be perceived as cheating or signal low ability. That concern is not unfounded: research on software engineers found that identical AI-assisted work received lower competence ratings when the engineer was female, a penalty that female engineers correctly anticipated. The fifth friction is trust, privacy, and risk. Women report stronger concerns about data security, are more likely to judge AI’s risks as exceeding its benefits, and are more sensitive to AI’s potential effects on employment. The researchers note that the gap from the first two frictions has shrunk as AI tools become more visible, practical use cases spread, and people see peers benefiting from AI. The remaining frictions, however, depend more on workplace design, norms, and trust, and are less likely to fade on their own.
Why This Matters
Generative AI is often described as a democratizing technology, but this research shows that access does not automatically produce equal use. If women are less likely to use AI tools, or use them less intensively, organizations may be leaving talent, efficiency, and innovation on the table. Business leaders and executives should be aware of the frictions that cause this inequality, and strive to create the conditions for equal participation: clear policies, hands-on training, encouragement, privacy safeguards, and workplace norms that reward thoughtful AI use rather than stigmatize it. Without that support, generative AI may reinforce existing inequalities instead of broadening opportunity.
References
[1] Cranney, Katelyn, Solène Delecourt, and Rembrand Koning, “Global Evidence on Gender Gaps and Generative AI Over Time,” Harvard Business School Working Paper No. 25-023 (May 2026): 2. https://dx.doi.org/10.2139/ssrn.6880085
[2] Cranney et al., “Global Evidence on Gender Gaps and Generative AI Over Time,” 11.
[3] Cranney et al., “Global Evidence on Gender Gaps and Generative AI Over Time,” 15.
Meet the Authors
Katelyn Cranney is pursuing a PhD in economics at Stanford University.
Solène Delecourt is an assistant professor in the Management and Organizations Group at the Haas School of Business.
Rembrand Koning is Mary V. and Mark A. Stevens Associate Professor of Business Administration at Harvard Business School, and the co-director and co-founder of the Tech for All lab at the HBS AI Institute.
Watch a video version of the Insight Article here.