AI Could Close the Girls-in-STEM Gap — If Parents Act Now

A ten-year-old girl takes apart a broken blender to see how the motor works. She is curious, confident, and completely unaware that she’s supposed to find this uninteresting. Two years later — right around seventh grade — something shifts. She starts saying she’s “not a math person.” She stops raising her hand in science class. She doesn’t know exactly when it happened. Neither do her parents.

That window between ages 10 and 12 is where one of the most well-documented and consequential gaps in education opens. And there is a real, research-supported argument that AI — handled correctly — is the best opportunity in a generation to close it before it does.

The Gap Is Real, It’s Getting Worse in Some Dimensions, and It Starts Early

The numbers on women in technical fields are not improving as fast as they should be, and in some areas they’re moving the wrong direction. Women currently represent only 12% of cloud computing workers globally and 26% of the AI workforce — the fields that are reshaping every other industry. In U.S. computer science bachelor’s degrees, women now earn 21% of degrees, down from a high of 37% in 1984. That is a regression, not stagnation. A field became more culturally male-coded over the past four decades, not less.

The World Economic Forum’s March 2025 analysis of AI and the gender STEM gap identified specific sectors where the gap is widest: machine learning, data engineering, and AI research. These are also the highest-growth technical sectors of the coming decade. The practical stakes of the gap are not abstract — they are wages, career trajectories, and who gets to shape the technologies that will affect everyone.

The gap doesn’t emerge at the university level. It opens much earlier. Research consistently points to ages 11 and 12 as the inflection point when girls begin self-selecting out of STEM pathways at disproportionate rates, often citing “not a math person” — a fixed-mindset identity statement that precedes the material actually becoming difficult. The algebra isn’t hard yet. The identity has already foreclosed the path.

This means that by the time a girl’s STEM trajectory looks like it’s heading somewhere, the key window for shaping it has already passed. And by the time she’s in high school, the cultural and psychological sediment around “who does engineering” is thick enough that interventions require substantially more effort.

What the Research Actually Says

The mechanism behind the gap is stereotype threat, first documented systematically by Claude Steele and Joshua Aronson in 1995 and replicated in hundreds of studies since. Stereotype threat refers to the performance-disrupting anxiety that occurs when a person is in a situation where they might confirm a negative stereotype about a group they belong to. Girls who are primed to think about their gender before a math test score significantly lower than girls who are not — even when the two groups have identical prior performance. When girls are told the test shows no gender differences in performance, the effect disappears entirely.

The implication is that the gender gap in STEM performance is not primarily about ability. It’s about the cognitive load imposed by stereotype threat in environments that feel identity-threatening. A girl in a classroom where “math is for boys” is an ambient cultural message is spending cognitive resources managing that threat that boys in the same room do not have to spend. She solves fewer problems. She appears to be worse at math. The stereotype looks confirmed.

AI tools in educational settings are creating a new and underappreciated version of this problem. Research published via phys.org in March 2026 documented that generative AI in classrooms can either reduce stereotype threat or amplify it, depending on how the AI is deployed. When AI tutors or problem sets use default examples drawn from male-coded domains — sports statistics, automotive mechanics, combat-based game scenarios — they quietly activate stereotype threat in female users. The AI is not neutral. It reflects the training data it was built on, which itself reflects the historical gender composition of STEM fields.

The same research found significant potential upside: when AI is deployed without these default male-coded framings — using gender-neutral or explicitly inclusive examples — it can meaningfully boost girls’ confidence in STEM. The tool’s effectiveness depends entirely on implementation choices that teachers and curriculum designers make, often without awareness of the research.

Girls Into Coding, a UK-based initiative, reported in 2025 that they had reached over 5,000 girls through hands-on AI and coding workshops. More than 90% of participants reported increased interest in STEM careers after those experiences. The hands-on, creating-something framing of those workshops is not incidental to the result — it is likely the mechanism.

Here is where the girls-in-STEM gap currently stands across education and career levels:

Stage	Girls/Women Participation	Trend	Notable Gap Drivers
Elementary school (K-5)	Roughly equal to boys	Stable	Gap not yet visible
Middle school (6-8)	Begins diverging around age 11-12	Widening	Identity formation, stereotype threat onset
High school AP STEM	~45% of AP Biology; ~25% of AP Computer Science	Mixed	CS gap is wider and growing
University — CS degrees	21% of bachelor’s degrees	Declining (from 37% in 1984)	Culture, belonging uncertainty
Tech workforce — general	~26%	Slowly improving	Retention issues, culture
AI/ML specifically	~26% of AI workforce	Slow improvement	Field is new but reproducing old patterns
Cloud computing	~12%	Stagnant	Heavily male-coded subfield

The pattern in this table reveals something important: the gap is not uniform. In biology and chemistry, girls participate at high rates through high school. The gap is specifically concentrated in computer science, engineering, and physics — the fields with the strongest cultural association with masculinity. AI sits at the intersection of computer science and math, which means it is inheriting both sets of cultural baggage.

But AI is also new. It doesn’t have 50 years of computer science classroom culture embedded in it yet. The parents and teachers engaging with 10-year-old girls around AI today are working with materials that don’t yet have hardened gender narratives attached to them. That window is narrow and closing.

What to Actually Do

Frame It as Making and Creating, Not Math and Computing

Research on stereotype threat and gender in STEM converges on a practical finding: girls’ performance in technical activities is equivalent to boys’ when the activity is framed as creative rather than mathematical. Making something — a working circuit, an animated character, a program that does something surprising — does not carry the same identity-threat load as “doing math.” The framing is doing work that the content cannot do on its own.

When introducing AI concepts to daughters, lead with what things are being made, not with what mathematics underlies them. “You can teach a computer to recognize your handwriting” lands differently than “we’re going to learn about classification algorithms.” Both are true. One activates curiosity; the other activates identity concerns.

Watch for Stereotype Threat Moments — and Intervene Specifically

Stereotype threat is activated by cues, not just overt statements. If your daughter’s coding class uses male-coded examples consistently, or if the AI tools she’s using surface images and scenarios that position men as the default technical actors, that is a real effect worth naming.

The research-backed intervention is specific: telling girls that a particular activity shows no gender difference in performance reliably eliminates the stereotype threat performance gap. This doesn’t have to be a lecture. It can be as simple as “girls and boys are equally good at this — actually, some of the best AI researchers in the world are women” offered conversationally before she sits down to work.

Find the Age 10-12 Window and Use It

The research is clear that this is the decisive window. A girl who is still curious and confident in STEM at age 12 is dramatically more likely to pursue STEM pathways through high school than one who has already internalized “not a math person.” The interventions that work before the identity shift are much lighter than what’s required after it.

This doesn’t mean forcing STEM. It means making sure STEM experiences available during this window are high-quality, framed as creative, and delivered in contexts where girls see women doing technical work as a normal thing. Camps, after-school programs, and parent-child projects all count.

Look for AI Tools That Are Tested for Bias

Not all AI tools used in educational settings have been evaluated for gender bias in their examples and outputs. The phys.org research makes clear that this is not a theoretical concern — stereotyped AI outputs produce measurable effects on girls’ confidence and performance. When evaluating AI tools or curricula for your daughter, ask whether the examples used are gender-neutral, whether the tool has been reviewed for stereotype-activating content, and whether there are women represented in the instructional examples and role models.

Introduce Role Models With Specificity

“There are lots of women in STEM” is not effective. Research on role model effects shows that specificity matters — girls benefit more from knowing about a specific woman who did a specific thing and faced specific obstacles than from general awareness that women can succeed in STEM. Name people. Tell their stories. The engineers who designed the systems your daughter uses every day are often women, and that fact is rarely surfaced.

For more on how an engineering mindset develops through failure and iteration, and why that process is available to all kids, see our pieces on developing an engineering mindset through failure and learning and future-proofing kids with AI and career skills.

What to Watch for Over the Next 3 Months

If your daughter is between 9 and 13, watch specifically for the identity-foreclosure signal: “I’m not a math person” or “science isn’t really my thing.” This statement often appears before there’s much academic evidence for it. It’s worth gently pushing back on it — not dismissively, but with specific evidence. “You figured out how to rewire that lamp last month — that’s exactly what electrical engineers do.”

Watch also for which kinds of technical activities she engages with voluntarily versus which ones feel like school. Voluntary engagement with creative technical activities — regardless of whether they’re labeled STEM — is the signal you’re looking for. A girl who is voluntarily making things, tinkering with things, or teaching herself things in technical domains is building exactly the self-concept that makes STEM accessible.

If your daughter is already in the “not a math person” phase, watch for contexts where that identity relaxes. When is she absorbed in a problem without worrying about whether she belongs? Those contexts are data about what kind of environment works for her, and they’re worth investing in and expanding.

Frequently Asked Questions

My daughter says she doesn’t like STEM. Should I push her?

Pressure and interest are different interventions. Pressure on a topic someone has decided they’re not good at tends to confirm the “not good at it” identity. Exposure to the topic through a different framing — particularly creative, hands-on framing — can change the relationship to the topic without any pressure. The goal is to make sure the “I don’t like STEM” conclusion isn’t based primarily on classroom experiences that activated stereotype threat, before assuming it reflects a genuine preference.

Are there specific AI tools that are better for girls?

The research isn’t yet at the point of recommending specific tools. What it does recommend is evaluating tools for the gender coding of their default examples. Tools that use a variety of examples and explicitly feature women as technical actors are preferable. Block-based coding environments tend to be lower in stereotype-threat load than traditional coding environments, which have stronger cultural associations with male programmers.

My daughter is 15 and already says she hates math. Is it too late?

The window is harder to work with at 15, but it is not closed. The research on stereotype threat elimination shows the effect is present and addressable even in adult women. What changes at 15 is that the identity is more entrenched and the intervention needs to be more sustained. Finding a context — a project, a summer program, a mentor relationship — where she experiences herself as technically capable, over a long enough period, can shift the identity. It takes more than a single experience.

What about boys who get excluded from “girls in STEM” programs?

The gender gap in participation and outcomes in these fields is real and documented. Programs designed to address that gap are responding to a structural problem, not creating one. Boys are not disadvantaged in STEM by the existence of programs designed to help girls access it — they currently dominate these fields by substantial margins.

How does this connect to AI specifically versus STEM generally?

AI is a special case because it’s new enough that cultural norms around who belongs are still forming. The girls who engage with AI tools, build with AI, and develop AI skills at ages 10-13 are among the first generation to do so. The cultural narrative about “who does AI” is being written right now, and the families and schools that actively include girls in that narrative are contributing to what the field looks like in 10 years.

---

About the author

Ricky Flores is the founder of HiWave Makers and an electrical engineer with 15+ years of experience building consumer technology at Apple, Samsung, and Texas Instruments. He writes about how kids learn to build, think, and create in a tech-saturated world. Read more at hiwavemakers.com.

Sources

1. phys.org. (March 2026). Generative AI has significant potential to empower girls in STEM. https://phys.org/news/2026-03-generative-ai-empower-girls-stem.html
2. World Economic Forum. (March 2025). AI and the gender STEM gap: Where the opportunity lies. https://www.weforum.org/stories/2025/03/ai-stem-women-gender-gap/
3. Girls Into Coding. (2025). Annual impact report: 5,000+ girls reached through AI and coding workshops. Girls Into Coding UK.
4. Steele, C. M., & Aronson, J. (1995). Stereotype threat and the intellectual test performance of African Americans. Journal of Personality and Social Psychology, 69(5), 797–811.
5. Spencer, S. J., Steele, C. M., & Quinn, D. M. (1999). Stereotype threat and women’s math performance. Journal of Experimental Social Psychology, 35(1), 4–28.
6. Master, A., Cheryan, S., & Meltzoff, A. N. (2016). Computing whether she belongs: Stereotypes undermine girls’ interest and sense of belonging in computer science. Journal of Educational Psychology, 108(3), 424–437.
7. National Science Foundation. (2025). Women, minorities, and persons with disabilities in science and engineering. NSF 25-309.
8. World Economic Forum. (2024). Global Gender Gap Report 2024. WEF.
9. Nguyen, H. H. D., & Ryan, A. M. (2008). Does stereotype threat affect test performance of minorities and women? A meta-analysis of experimental evidence. Journal of Applied Psychology, 93(6), 1314–1334.