Can AI Actually Close the Gender Gap in Women's Health?

You've probably seen AI pitched as the thing that's finally going to fix women's health — close the diagnosis gap, catch what doctors miss, make research move faster for conditions that have been underfunded for decades. Here's the honest answer from the people actually studying it: not automatically, and not without a lot of care. A group of seven researchers across gender studies, biomedical engineering, and AI — including Dr. Annelies Kleinherenbrink of Radboud University — published a paper this month arguing that simply pointing artificial intelligence at women's health doesn't close the gap on its own. Done carelessly, it can make things worse.

Can AI actually close the gender gap in women's health?

Not by default. The paper, "Beyond the hype of AI as a panacea for women's health," published in npj Women's Health in August 2026, lays out three interconnected problems. First, medical research has long relied on thin, binary categories of sex and gender that miss how bodies actually work. Second, when AI models train on that same incomplete data, they don't correct the bias — they scale it. Third, tools marketed as "empowering" women can end up expanding surveillance of their health data without addressing the structural reasons the gap existed in the first place, like which conditions get research funding and which symptoms get taken seriously in the exam room.

What the researchers actually found

Kleinherenbrink and her co-authors, based across Radboud University, MIT, the University of Edinburgh, Swarthmore College, UC Berkeley, and the University of Lausanne, aren't arguing against using AI in women's health. They're arguing against using it uncritically. "Equity needs careful evaluation at every stage of the AI pipeline, from problem definition to data collection and use," the authors write — meaning the question can't just be "does this tool work," it has to be "who decided what counts as working, and on whose data." That distinction matters in a field where women's health has historically been treated as a niche category instead of half the population. Feeding a model more of that same lopsided data doesn't erase decades of medicine built around a default male patient. It can just make that default faster, and harder to question.

Why this matters for you

This isn't abstract if you've ever sat in an appointment feeling like your symptoms were being minimized, or filled out a health app that asked questions clearly built with someone else in mind. We've written before about why so many young women don't trust the health care system, and this research is the missing piece — that mistrust isn't paranoia, it's built on a real pattern, and AI trained on the same pattern will repeat it unless someone actively designs against it. It also connects to what we've said about what happens when "fixing" AI bias goes sideways: a quick patch on a biased system can look like progress while the underlying data problem stays untouched. If you use health apps, symptom checkers, or an AI chatbot to ask questions about your body, you're already part of this story, whether or not the product you're using was built with that history in mind.

One thing to try today

Next time a health app or AI tool gives you an answer about your body that doesn't quite fit, don't assume you're the outlier. Ask a real provider, or look up whether the tool discloses what data it was trained on. It's a small habit, but it puts you back in the loop instead of just trusting the output — and it's exactly the kind of scrutiny researchers like Kleinherenbrink are asking the whole industry to build in from the start. It ties into a bigger question we keep coming back to about who actually gets to build the AI systems shaping women's lives.

Quote to sit with

"Equity needs careful evaluation at every stage of the AI pipeline, from problem definition to data collection and use." — Dr. Annelies Kleinherenbrink

💌 Sarah