The AI Jobs Gender Gap Is Widest Without a Degree

The same job can be exposed to artificial intelligence without exposing every worker to the same future. A new analysis of American employment suggests that distinction is becoming harder to ignore. In occupations where generative AI could assist with more of the work, men’s employment has grown faster than women’s. The difference is sharpest among people without a four-year college degree.

That is not proof that AI has displaced women. It is a warning about how technological change can meet an unequal labour market. If employers automate the paperwork, customer correspondence and information-handling tasks concentrated in some jobs, who gets the chance to move into the more valuable work that remains?

A striking divide in the latest US employment data

On 22 October 2026, Pew Research Center published an analysis of 480 US occupation groups, using government employment data from 2017-2019 and January 2023-June 2026. Researchers compared employment growth for men and women within occupations and tested whether their differences were greater in occupations more exposed to generative AI.¹

For each one-point increase on Pew’s occupation-level AI-exposure scale, men’s employment grew by an estimated 6.3%, while women’s grew by just 0.2%. The scale runs from zero to 3.5. These figures describe the relationship between exposure and employment growth across occupations and periods. They are not annual growth rates, the percentage of workers replaced by AI, or the effect of an individual using a chatbot.¹

Education reveals the more consequential split. Among workers with a high school diploma or less, each additional point of occupational AI exposure was associated with an 8.8% increase in men’s employment and a 9.5% decrease in women’s. For people with some college education but no bachelor’s degree, the corresponding figures were 3.6% for men and minus 6% for women. Among bachelor’s degree holders and above, Pew found no statistically significant employment change associated with increasing exposure for either sex.¹

This is not a story in which all women lose and all men win. Men without degrees are heavily concentrated in less-exposed occupations such as transport, construction and freight handling; relatively few hold the more exposed jobs in which the study detected their gains. Women without degrees are more frequently employed in secretarial, reception and customer-service roles, where writing, record-keeping and information processing can be assisted by AI. The distribution of occupations matters as much as the technology.

Exposure is a measure of tasks, not a count of layoffs

Pew’s measure is built from the US Department of Labor’s O*NET descriptions of work activities. Researchers classified 16 activities, including analysing information, writing, advising and documentation, as highly exposed to generative AI. Nine activities requiring physical or in-person work, such as repairing equipment and moving objects, were classified as less exposed. The remaining 16 activities were excluded from the exposure calculation.¹²

That design is useful for spotting where AI might alter work. It cannot show whether a particular employer bought an AI tool, whether it replaced an employee, or whether output quality improved. An occupation can score highly because its tasks are technically assistable even when privacy rules, customers, cost or management practices prevent automation. Conversely, an occupation with relatively low exposure today may still change through robotics or other technologies.

Pew excluded 2020-2022 because the pandemic distorted employment patterns. Its technical appendix reports roughly four million survey records, with regression results weighted by pre-pandemic occupational employment and standard errors clustered by occupation. Four of five exposure measures produced a gender gap of roughly five to seven percent per standard deviation; the image-generation-specific measure was the exception.² That robustness makes the association worth taking seriously, but it does not turn an observational comparison into a causal experiment.

Hiring freezes, shifts in consumer demand, industry restructuring, educational pathways and changes in who enters or leaves the labour force could all contribute. A business leader cannot responsibly translate these figures into a claim that AI eliminated a given percentage of women’s jobs. The stronger conclusion is that existing employment inequalities deserve attention when companies redesign AI-exposed work.

Why the opportunity gap may matter more than the automation rate

Consider a hypothetical regional charity with a small administrative team. One employee handles correspondence, appointment scheduling and basic reporting. AI might shorten the time needed to draft replies and assemble routine summaries. The organization could respond by cutting hours, by increasing service volume, or by giving the employee more responsibility for sensitive cases, donor relationships and quality assurance. The same technology creates three very different employment and learning outcomes.

The decisive question is whether the employee has access to training, trusted tools and supervised opportunities to take on the higher-value work. A colleague who already holds a professional qualification may be invited to redesign the process. Someone hired for clerical duties may be judged only on the time saved. Without deliberate progression, automation can remove the tasks on which a worker’s current role is valued while leaving the next role out of reach.

That possibility also explains why “everyone needs AI skills” is an inadequate workforce strategy. A receptionist does not necessarily need to become a machine-learning engineer. They may need to learn how to verify AI-assisted messages, recognise confidentiality risks, resolve exceptional cases and communicate clearly with people whose needs do not fit a template. Those capabilities should be recognised as real work, with time to learn them and a path to more responsibility.

The international evidence supports concern, not a universal prediction

The International Labour Organization’s 2025 global occupational-exposure index found that in high-income countries, 9.6% of women’s employment was in occupations with the highest potential for generative-AI automation, compared with 3.5% of men’s employment. Clerical work stood out. The ILO also stressed that exposure generally signals likely transformation of tasks rather than automatic disappearance of entire jobs.³ These global figures measure potential exposure, not the US employment changes measured by Pew, and should not be treated as the same statistic.

A separate Pew survey published on 16 October 2026 adds another layer. In February, 47% of US women and 50% of men said they had ever used AI chatbots, a relatively small gap in access at that broad level. Yet daily use was more common among men, 27% versus 20%, and women were less likely to say chatbots helped their productivity.⁴ These are self-reported attitudes and behaviours, not a direct explanation of the employment gap. They do, however, challenge the idea that merely offering access to a tool is equivalent to giving everyone an equal opportunity to benefit.

Neither study establishes that a particular training programme would close the gap. Evidence about which interventions work, for whom and over what period is still needed. Nor can US employment patterns be automatically exported to Europe, where occupational structures, labour law, childcare arrangements and vocational pathways differ. The ILO’s international findings justify asking similar questions elsewhere, not assuming identical answers.

What a small employer can test before jobs change

The practical response is not to assign a gender-based risk score to employees. It is to inspect how work and opportunity are distributed. Start with the tasks most likely to change, then identify who performs them, who has been trained on approved AI tools, who checks outputs, and who is being offered the new responsibilities that follow. Review the findings by role and employment arrangement, and examine whether participation in training or advancement differs across groups without making automated employment decisions about individuals.

Next, run a supervised pilot in which people doing routine work help design the improved process. Measure time saved alongside error rates, customer experience, escalation workload and the skills employees can demonstrate without assistance. Give workers an explicit development objective rather than only a productivity target. In a small charity, that could mean moving from standard correspondence to handling complex client enquiries; in a service company, from logging requests to diagnosing and resolving exceptions.

Finally, ask who benefits from the time released. Is it reinvested in better service, learning and internal mobility, or captured solely as a reduction in labour costs? Both may be legitimate business choices, but they have different consequences for the future workforce. If only the most credentialed employees are offered advancement, an AI efficiency project may quietly reinforce a divide it never set out to create.

The latest data do not show that AI has caused a gender employment crisis. They show that the people best placed to benefit from AI-related change may not be the same people whose existing tasks are most exposed. For leaders, the test is no longer simply whether automation works. It is whether the route from a changing job to a better one is genuinely open.

How hilighter can help

When AI changes tasks unevenly, organizations need visible roles, skills and progression - not only efficiency metrics. hilighter helps teams map skills by role, compare people with requirements through Skills & Talent Insights, recognize evidence of capability in Skills Evidence & Recognition, set personalized growth paths, and explore an indicative view of AI’s potential impact on jobs. A free plan covers up to three roles.

Sources

1. Smith, N., Pew Research Center, Men Are Experiencing More Job Growth Than Women in AI-Exposed Occupations, 22 October 2026.
2. Pew Research Center, Technical Appendix, 22 October 2026.
3. International Labour Organization, Generative AI at Work: What It Means for Jobs in Europe and Beyond, 29 September 2025.
4. Anderson, M. and Park, E., Pew Research Center, The Gender Gap in AI, 16 October 2026.

Sources last reviewed: 27 October 2026. Pew figures are observational associations between occupational AI exposure and employment growth, not causal estimates of AI-driven layoffs. ILO figures measure potential task exposure, not the same US employment changes.