Two colleagues face the same task with the same tools. One opens a generative AI assistant and drafts the proposal in minutes; the other hesitates, worried about getting it wrong, and works through it the slow way. Repeated across millions of workplaces, that quiet difference tends to break down along gender lines — and the resulting AI adoption gap has become one of the most consequential, and least discussed, forces shaping who benefits from the technology and who is left behind.
The numbers behind the AI adoption gap
A Harvard Business School working paper that pooled 18 separate studies — covering more than 143,000 people across 25 countries — estimated that women have roughly 22% lower odds of using generative AI than men. The pattern held across countries, sectors, occupations, and specific tools, and persisted even in field experiments where access and training were equalised, which suggests that access alone does not explain it.
Survey work drawing on the U.S. Survey of Consumer Expectations points in the same direction, with about 37% of women versus 50% of men reporting they had used generative AI. Estimates of the gap’s size vary with how the question is asked: the OECD’s own January 2026 cross-country figures put the difference far smaller, at around four percentage points. Policymakers, including the OECD, have begun treating the divide as an economic question rather than a curiosity.
The confidence penalty
The gap appears to be driven less by ability than by perception. Self-assessed knowledge is among the strongest predictors of whether someone uses these tools, and women, on average, rate their own AI competence lower even at similar skill levels. Privacy concerns and differing judgements about the balance of opportunity and risk explain a further share of the difference. Social signalling matters too: in some settings, using AI is quietly read as a shortcut, and people who fear being judged for “not doing it themselves” are slower to adopt.
Why this matters for small businesses
For a small business, an internal adoption gap is a productivity gap. When part of a team routinely uses AI to draft, summarise, and analyse while another part does not, output and turnaround diverge — and that divergence can be mistaken for a difference in talent. It also compounds an existing problem: women already face a steep funding gap when building AI startups, and slower hands-on adoption widens the experience divide that investors and clients later read as a lack of expertise.
How to close the gap within an organization
Make experimentation safe and visible
People adopt what they see leaders use openly. When a manager — particularly a senior woman — demonstrates how she used AI to draft something, it signals that the tool represents competence rather than cheating.
Train for the workflow, not the hype
Generic “intro to AI” sessions rarely move the needle. Showing people the two or three tasks in their actual role that AI can shorten this week builds confidence faster than demonstrations of impressive but irrelevant tricks.
Reward outcomes, not tool theatrics
Judging the proposal, the campaign, or the analysis — rather than whether the work looked effortful by hand — removes the competence penalty, which is among the most effective ways to narrow the gap.
Mind data and defaults
Clear guidance on what may be entered into which tools removes a real source of hesitation, especially where privacy worries are doing much of the work. Sensible defaults let cautious staff participate without weighing a policy decision every time.
Limitations and what to watch
The size of the AI adoption gap is genuinely contested. Meta-analyses of behavioural studies report a large gap, while some national surveys report a small one, so headline percentages should be read as a range rather than a single fact. Much of the underlying research measures self-reported use, which can diverge from actual use, and adoption is moving quickly enough that any single figure dates fast. The broad direction — a persistent, confidence-linked gap that access alone does not close — is better supported than any precise number.
A gap worth closing
The AI adoption gap is not a story about who is capable. It is a story about who feels permitted to experiment, and the organizations that make that permission explicit will capture both the productivity and the talent that others leave on the table. For related context, see this site’s coverage of how employees are quietly learning AI on the job and how small businesses can stay resilient as AI tools change. The underlying research is available in the Harvard Business School working paper Global Evidence on Gender Gaps and Generative AI and the OECD’s analysis of AI and the gender gap.
