Twenty-nine percent. That is the share of female-dominated occupations exposed to generative AI, according to International Labour Organization data released this year — against just 16% of male-dominated ones. Nearly double. The gap widens further at the sharp end: 16% of women’s jobs sit in the highest automation-risk category, compared with 3% of men’s. Understanding AI job exposure for women means starting from an uncomfortable fact — the technology is not landing evenly, and it never was going to.
Why the numbers break this way
The explanation is not that AI has a bias against women. It is that AI is very good at exactly the work women were historically funnelled into. Clerical, administrative and business-support roles — secretarial work, payroll, scheduling, data entry, records management — face the highest generative-AI exposure of any occupational category, because their tasks are routine, text-heavy and rule-bound. They are also overwhelmingly done by women.
In other words, decades of occupational segregation have quietly become an automation liability. The ILO’s researchers make this link explicitly: the exposure gap is a downstream consequence of who was steered into which jobs. A UN analysis puts it starkly — roles traditionally held by women are close to three times more likely to be transformed by AI than those held by men, 9.6% against 3.5%. In the US, women make up 57% of workers in roles most likely to be affected.
Exposure is not the same as elimination
An important distinction gets lost in the headlines. The ILO frames its findings around transformation, not replacement. “Exposed” means a meaningful share of the tasks in a role can be performed or accelerated by generative AI. It does not mean the job disappears. Most exposed roles change shape: the routine core shrinks, and what remains is the judgement, coordination, relationship and exception-handling work that models still handle poorly.
That distinction is where the opportunity sits — but only for people who see the change coming. A role that is 40% automatable becomes a role where the human does the other 60% at greater scale. The person who understands the tools directs them. The person who does not gets measured against someone who does.
The exposure gap meets the adoption gap
Here is the cruel arithmetic. Women hold the jobs most exposed to generative AI, and women are simultaneously using generative AI at work at a rate roughly 25% lower than men. Lean In’s research finds this is not simple reluctance: women receive less managerial encouragement to use AI, and are more likely to expect their AI-assisted work to be judged rather than rewarded. The hesitation is, in its way, a rational read of the workplace.
But rational or not, the compounding effect is brutal. The people with the most to lose from automation are adopting the tools most slowly. We covered the business-ownership side of this in our look at why women-owned businesses are falling behind on AI, and the capital side in why women AI founders still fight for funding. The pattern rhymes across all three: exposure is high, access is low, and the two are moving in opposite directions.
What actually closes the gap
The interventions that work are unglamorous and specific:
- Explicit permission and encouragement. The single strongest predictor of whether someone uses AI at work is whether their manager visibly expects them to. Sanction it out loud.
- Credit, not suspicion. If AI-assisted output is treated as cheating rather than competence, the most-scrutinised employees will opt out. Set the norm deliberately.
- Training inside work hours. “Learn it in your own time” is a tax paid disproportionately by people already carrying more unpaid labour at home.
- Move up the task chain. The durable roles are the ones that direct AI rather than compete with it: process design, quality control, exception handling, client relationships, AI oversight and governance.
That last point matters most, and it is where the ceiling is currently lowest — women hold only about 14% of AI executive roles and 18% of AI research positions. The governance and oversight functions being built right now are the ones with real durability, which is exactly why women taking the governance seat is more than a representation story. It is a career-survival story.
The window is open, briefly
Roughly 80% of the global workforce will need new skills by 2027 to stay competitive, on the World Economic Forum’s estimate. That is a daunting figure, but it is also a levelling one: when nearly everyone has to retrain, incumbency counts for less than it usually does.
The women in the most-exposed roles already know their work better than any consultant does — they know which parts are drudgery and which parts quietly hold the organisation together. That knowledge is the raw material for redesigning the role rather than being redesigned out of it. The exposure data is not a prophecy. It is a warning with a lead time, and lead time only helps the people who use it.
