When I was designing the cover of my book, Ambitious Mother, I tried to get a little help from artificial intelligence. We had worked hard on the visuals with the design team, and after many drafts I just wanted to see what a different color font would look like, without asking them for more changes. I put the photo into AI and prompted it to show me the exact same cover, but with green text instead of red.
Imagine my surprise when the image that came back was the same cover, in the new color, but instead of Ambitious Mother, the book had been renamed Ambitious Father. And the author’s name had changed from Anne Welsh to John Welsh.
I was shocked and also not.
On one hand, I had a healthy skepticism of AI. We knew it would “hallucinate” articles. I knew to constantly double-check the research, the numbers, the citations, and more. I did that diligently.
At the same time, I had made a very simple request and the response felt like much more than a mistake. It felt like a statement: AI was SO unfamiliar with the idea of these two words being placed next to each other, it had to make it a new set of words to finish the prompt. And then changed the author’s gender to match.
We have an assumption that computers are not biased because they are based on data and therefore somehow objective. And yet, AI is trained on what already exists, and what exists is biased. Whose stories are told in the history books? Whose stories are represented in the data? Who do we study and how to do we interpret that data?
The real danger is that people don’t always know what bias is baked in, and we aren’t always asking the right questions.
An Old Double Bind, Dressed up in New Tech
I’ve been thinking about that story again as we talk more and more about the gender gap in AI use. More often than not, we talk about gender gaps in leadership as if they are a woman’s problem to solve. We tell women: Speak up. Be more confident. Negotiate. Advocate for yourself.
The problem with this advice is that behaviors are not treated the same in men and women.
We have known this for a long time although we don’t always talk about it. For example, research on the leadership double bind has found that women need to demonstrate the strength associated with effective leadership while also demonstrating the warmth we expect from women. Women who are assertive enough to be seen as competent can then be penalized for not being warm or likable enough.
I see this play out with my clients in many ways. One client was an attending at a major health center. At multiple points in her training she watched male colleagues receive praise for their amazing bedside manner when they asked a simple question that involved any degree of empathy. Meanwhile, female colleagues consistently demonstrated empathy and excellent bedside manner, and it was simply expected. In contrast, when women veer from that expectation, they receive sharp criticism, while men are exempt.
Now I see us doing a version of the same thing with AI.
What the Research Shows
There is an AI gender gap. According to recent research from Lean In, men are 22% more likely than women to say they use AI daily or constantly at work. This is important to notice and fix. AI is going to be part of how we work, and addressing this is important.
But we keep talking about the gap as though the solution is simply getting women to use AI more. That ignores the environment in which we are asking them to adopt it.
In the same Lean In research, men were 23% more likely than women to say their managers encouraged them to use AI, and 27% more likely to say they had been praised for using it. Meanwhile, women were 32% more likely to worry that using AI would be perceived as cheating.
And women may be correctly assessing the situation as more fraught for them.
In a recent study of more than 1,000 software engineers, researchers had participants evaluate identical code while changing whether they were told it had been produced with AI and whether the engineer was a man or a woman. When people thought AI had been used, everyone took a competence hit. But women took a bigger one. The competence penalty for women was more than twice the penalty for men.
In this study, the work didn’t change. What changed was the evaluator’s knowledge that AI had helped produce it. And while everyone took a hit to perceived competence, it was significantly different for men and women.
In a follow-up survey in the same paper, researchers also found that people who anticipated a greater competence penalty were slower to adopt AI, and that relationship was stronger among women.
Essentially, women accurately perceive that they will be judged more for using AI and are therefore less likely to adopt it. And this is where everything feels familiar. Women are told just do X, without the context that X is perceived differently for men and women.
Women were told to negotiate more. Later research found that women were asking for raises and promotions as often as men but were less likely to get them, while other studies have found that women can be penalized for initiating negotiations.
And now we’re saying: Just use AI. Women need to learn it. They need to catch up.
But when they do, they may be perceived differently.
We also have to address the issue of invisible labor and the leisure gap here. Women have less discretionary time. OECD data have consistently found that women spend more time on unpaid work and less time on leisure than men. This means that outside work, they simply have less time to address the steep learning curve that high-level AI use requires.
There are also differences in how we value the time that women do put into AI learning. In the women’s leadership groups that I lead, I have heard from multiple women who are taking on AI adoption roles in their organizations without having that work compensated or even really named as work. They are asked to be on the AI committee, to help other people figure it out, or to think through some of the ethical and people questions around AI. And in multiple cases, I have watched those roles become more formalized only to have the formal roles go to men.
I’ve written before about the problem of invisible labor at work, so I won’t rehash it here. But I do think we have to ask a basic question: If women are spending their time doing the invisible work of AI adoption for everyone else, or doing all of the other invisible work that still disproportionately falls to women, when exactly are they supposed to find the extra time to learn and experiment on their own? And why aren’t we giving them credit for any of it?
Baked-in Bias
And then we come back to the technology itself.
Circling back to the idea of warmth and competence we started with, we see the same perception reiterated by technology. Researchers compared human and AI-generated judgments using the classic dimensions of warmth and competence. In ChatGPT-generated ratings, women were portrayed as warmer and less competent than men. Even more strikingly, the gender differences were larger in the AI-generated ratings and images than in the comparable human-generated material.
And that cover swap I told you about? In a 2026 study out of Germany, researchers generated 1,344 images with various prompts. When the prompt was to generate an image of “a person” or “a warm person” the AI was more likely to show women. When the prompt was “a competent person” the image was more likely to be a man.
This shows up in specific work contexts too. A 2025 study published in Nature asked ChatGPT to generate fictional résumés across 54 occupations. When the applicant had a woman’s name, ChatGPT generated applicants who were younger, had graduated more recently, and had less relevant experience than when the applicant had a man’s name.
That doesn’t mean AI is going to take your résumé and suddenly erase years of your experience, but it does show that we cannot assume objectivity simply because the information came from a computer. AI makes assumptions about women’s competence and experience based on baked-in bias. We have to work against that in our workplaces and in our own use of AI.
What Companies Can Do
As with all things, there are individual and systemic contributions and solutions here. And once again, we are taking a systemic issue and offering only individual solutions: Women just need to catch up with AI.
Instead, we need to see both/and.
Yes, women need the skills to use AI and we need to address this gap. But, we need to address the gap by addressing the bias, not just the women. We can think about it in four categories:
1. Encouragement
Organizations need to question how they approach AI and ask themselves: Who are you encouraging to use AI? Who has access to the tools and the training? Who are you praising when they do? Who is doing the work of AI adoption, and who eventually gets recognized as the expert?
2. Time
Learning any new tool takes time. Time to play, to make mistakes, to experiment. If AI is important, then make sure the women also have time to learn it. It cannot be left to “spare time” because we know that women have less of that.
3. Visibility
If someone is spending time on AI work for your organization—teaching, evaluating, learning—make sure that it all counts. Name it, include it in promotion conversations, and make sure they get credit even before it’s formalized.
4. Evaluation
We need to pay attention to how we evaluate work. Standardize what constitutes appropriate AI use and how the finished product gets evaluated. If knowing that someone used AI changes how competent you think they are, would you make the same judgment if the person doing the work were someone else?
Of course, we also need to bring that same scrutiny to the AI tools themselves, particularly when they are being used to make decisions about people. If AI is helping screen candidates, summarize performance feedback, or make recommendations about talent, we cannot assume that its output is objective simply because a computer produced it. The emerging research on gender, warmth, competence, and experience gives us good reason to keep asking questions.
If we want women to adopt AI at the same rate as men, we have to address the differences in time, opportunity, perception, judgment, and recognition that make adoption a different proposition for women in the first place.
Otherwise, we’re once again identifying a gender gap and telling women to fix it.
