Women are being pushed out of work by AI faster than men are, and this is measurable. It shows up in how their AI-assisted work gets graded, in which jobs sit closest to the automation trapdoor, and in a wariness toward the tools that keeps getting misread as falling behind.
The Competence Penalty
Start with how the work gets judged. In a study published in Harvard Business Review, researchers ran a pre-registered experiment with more than 1,000 engineers who evaluated the same Python code, described to some as handwritten and to others as AI-assisted. When people believed a woman had used AI to write it, their confidence in her skill fell to roughly half of what a man received for identical code. The code never changed. The verdict did. So women compete with men for the same jobs while being graded on a harsher curve the moment they touch the tool.
First in Line
The pressure is also structural. The International Labour Organization finds that in high-income countries, 9.6% of women hold jobs at high risk of automation, against just 3.5% of men. That lines up with where women actually work. The white-collar, office-based world is around 70% female, and offices are exactly what AI is absorbing first: writing, data processing, scheduling, the everyday text tasks that large language models already handle at scale. Men lean toward physical roles, which stay harder to automate, at least for now.
Access, or Something Deeper?
You might guess women use AI less simply because they have less access to it. The data complicates that. Between November 2022 and May 2024, women made up just 42% of ChatGPT's roughly 200 million average monthly website users, and only 27% of its app downloads. One study cut the other way: a Boston Consulting Group survey of San Francisco tech workers found women 3% more likely to use AI than men. "Women in tech may have had more exposure to these tools and are more comfortable using them," notes researcher Rembrand Koning, hinting that familiarity, not gender, is the real lever.
To test it, Koning and colleagues ran a field experiment with about 17,000 entrepreneurs in Kenya who were invited and told to use ChatGPT. Even with access handed to everyone equally, women were about 13% less likely to try it. So access was not the wall.
Why the Distrust Runs Deep
Some of it loops back to that competence penalty. If simply opening the tool invites suspicion about your honesty, the tool stops feeling like help and starts feeling like a setup. But the caution is also principled. Women are 29% more likely than men to question AI's accuracy and 38% more likely to raise ethical concerns about it. Add in firsthand run-ins with hallucinations, the confident nonsense these systems sometimes produce, and the skepticism looks less like fear and more like plain judgment.
The Real Work Ahead
Stack it up and you get compounding disadvantage. Greater exposure to AI-driven job loss. Harsher criticism for using the tool at all. And sensible caution about reliability, read as resistance. For anyone hiring or building teams right now, that is a live problem: you cannot mandate AI productivity with one hand and quietly dock half your staff for using it with the other.
Handing women more logins will not close this. Access does not undo decades of being kept out of the room. The harder fix is the double standard underneath, who gets trusted with these tools and who has to prove themselves twice. AI did not invent that imbalance. It showed up and inherited it, the way it inherits everything else about the world it learned from. Until that older unfairness gets faced directly, no amount of access or training will make the field level.
