AI Widens the Gender Gap  🔗 **Source:** [AI Widens the Gender Gap](https://leaddev.com/ai/ai-widens-the-gender-gap) > [!note]+ Document Notes > tags: [ai-adoption, gender-equity, engineering-management, workplace-bias, psychological-safety] > author: LeadDev (byline not captured) > source: LeadDev > reference: https://leaddev.com/ai/ai-widens-the-gender-gap > date created: 2026-09-21 > > > [!info] Generated analysis — not the author's words. Sections below are machine-written > > from the captured source. Only text inside `> [!quote]` blocks is verbatim from the author. > > Do not quote or attribute any other line to the named author. > > # 🏷️ Title: AI widens the gender gap > <!-- deep-read:sections start --> ## 🧭 Table of Contents - [📝 Summary](#-summary) · [📖 Main Content](#-main-content) · [🗒️ Notes](#️-notes) · [📚 References](#-references) · [🗣️ Paraphrase](#️-paraphrase) · [📰 Newsletter](#-newsletter) · [✅ Action Items](#-action-items) · [🧪 Hypotheses](#-hypotheses) · [📊 Methodology](#-methodology) · [❓ Questions](#-questions) · [🔮 Future Research](#-future-research) · [💡 Implications](#-implications) · [🧠 Learning Styles Assessment](#-learning-styles-assessment) · [📊 Accessibility Assessment](#-accessibility-assessment) · [🌍 DEIB Assessment](#-deib-assessment) · [✅ Logical Validity](#-logical-validity) · [🛡️ DARVO Analysis](#️-darvo-analysis) · [🎯 Conclusion](#-conclusion) · [🧱 Refactor Appendix](#-refactor-appendix) · [🧰 Template Library](#-template-library) ## 🔗 References (in-page) - Source: https://leaddev.com/ai/ai-widens-the-gender-gap — LeadDev, 2026-09-21 (author byline not captured in the Reader copy) - Harvard Business School paper, "Global Evidence on Gender Gaps and Generative AI Over Time": https://www.hbs.edu/faculty/Pages/item.aspx?num=66548 - Lean In data: https://leanin.org/articles/findings/ai-women-gender-gap-data/ - Harvard Business Review, "The hidden penalty of using AI at work": https://hbr.org/2025/08/research-the-hidden-penalty-of-using-ai-at-work - Deloitte research (link truncated in capture): https://www.deloitte.com/us/en/insights/industry/technology/technology-media-and-telecom-predic… ## 📝 Summary **What happened.** LeadDev argues that AI is carrying existing workplace gender inequality into a new technology. Women adopt AI at work more slowly than men, the article says, because they get less encouragement and training — and when they do use it, they're judged more harshly. **The evidence it cites.** - **Lean In** (1,000+ Americans): women are 22% less likely than men to use AI regularly at work; men are more likely to say managers encouraged them and 27% more likely to say managers praised them for it. - **Deloitte:** 93% of men vs 84% of women say their company actively encourages AI use; 91% of men vs 72% of women have had formal AI training. - **HBR** (1,000+ engineers): identical code was rated 9% worse when reviewers believed AI helped write it. When the coder's gender was revealed, women who used AI were rated 13% less competent, men 6%. Among male engineers who don't use AI themselves, the penalty for women reached 26%. - **HBS paper:** AI can boost productivity while amplifying inequality; women may keep lagging because of less training and institutional support. **The mechanism.** Hilary DeCesare (The reLaunch Company) describes a feedback loop: a man tries AI, is encouraged and praised, and comes to see himself as a capable user; a woman gets less encouragement and less credit, so the evidence never lands. Because AI fluency is becoming a proxy for speed, judgment, and leadership potential, a usage gap can become a career gap. Jossie Haines, a former VP of software, adds that most AI enablement treats everyone the same and reaches only people who'd have got there anyway. **The fix.** Stop just pushing adoption; remove the penalty for using AI; measure the actual gap instead of assuming your culture is the exception. **Why it matters.** Engineering managers are rolling out AI now. This piece gives them concrete, sourced reasons to check their own data. **Problems.** The closing quote is attributed to "Decrease," apparently a typo for DeCesare, right after a paragraph about Haines — so it's unclear who said it. The key-takeaways box says the gap is "not a skills gap," but none of the cited data measures skill; it shows differences in encouragement, training, and evaluation. And "22% less likely to use AI regularly" becomes "use AI at work 22% less" in the takeaways, which isn't the same claim. **Bottom line.** Well-sourced and useful, with four linked studies behind its main claims. Read the numbers in the body, not the takeaways box, and check the attribution of the closing advice before quoting it. ## 📖 Main Content The article opens with the thesis: AI is changing work, but old inequalities are coming along. It cites the HBS paper's finding that generative AI can raise productivity while amplifying existing inequalities — reinforcing disadvantages some claim it will disrupt. Lean In's survey of more than 1,000 Americans supplies the headline number: women are 22% less likely than men to use AI regularly at work. DeCesare says the data confirms what many have sensed, and that because AI now drives productivity, visibility, and advancement, the usage gap may decide who gets ahead. "New tech, same story" explains the gap as unequal conditions, not unequal interest. Per the HBS paper, women may lag because they get less training and institutional support. Lean In found men more likely to say their managers encouraged them and 27% more likely to report praise. Deloitte found the same pattern: 93% of men vs 84% of women say their company encourages AI use, and 91% vs 72% have had formal training. Haines argues enablement should meet people where they are — the skeptic needs something different from the daily user — and that working that out is a core management job. "The AI competence penalty" turns from access to judgment. Women face more scrutiny when they use AI, which creates fear of using it. The HBR study of more than 1,000 engineers found identical code rated 9% worse when reviewers thought AI helped; women who used AI were rated 13% less competent against 6% for men; and male engineers who don't use AI rated women who did 26% less competent. DeCesare describes the resulting loop and adds that for women the question isn't "will this help me?" but "what does it cost me if this goes wrong in front of the wrong person?" When AI fluency stands in for speed, judgment, and leadership, women who use it less may miss high-profile projects and promotions. "Fixing the gender gap" argues adoption is about who feels permitted to experiment, who gets rewarded, and who risks their reputation by getting it wrong. Haines says organizations should stop simply pushing adoption and remove the career penalty. The final quote — "Start by measuring the exact gap instead of assuming your culture is the exception" — is attributed to "Decrease." ### 🗒️ Notes - **Takeaways box:** women use AI 22% less (body: 22% *less likely to use it regularly*); "not a skills gap"; women rated 13% less competent vs 6% for men; measure adoption directly. - **Lean In:** 1,000+ Americans; 22% less likely to use regularly; men more likely to report manager encouragement; 27% more likely to report praise. - **Deloitte:** encouragement 93% (men) vs 84% (women); formal training 91% vs 72%. - **HBR:** 1,000+ engineers; AI-assisted code rated 9% worse; competence penalty 13% (women) vs 6% (men); 26% from male non-users rating women. - **HBS paper:** productivity gains alongside amplified inequality. - **DeCesare:** the encouragement-and-praise feedback loop; "what does it cost me if this goes wrong in front of the wrong person?" - **Haines:** one-size enablement reaches only early adopters; the job is to meet people where they are; remove the penalty. - **Defect:** "Decrease says" — likely DeCesare; follows a Haines paragraph, so the attribution is ambiguous. ## 📚 References Research linked in the article: - Harvard Business School, "Global Evidence on Gender Gaps and Generative AI Over Time" — https://www.hbs.edu/faculty/Pages/item.aspx?num=66548 - Lean In, AI and the gender gap data — https://leanin.org/articles/findings/ai-women-gender-gap-data/ - Deloitte Insights, TMT predictions (truncated in capture) — https://www.deloitte.com/us/en/insights/industry/technology/technology-media-and-telecom-predic… - Harvard Business Review, "Research: The Hidden Penalty of Using AI at Work" (2025-08) — https://hbr.org/2025/08/research-the-hidden-penalty-of-using-ai-at-work LeadDev's related coverage: - Are you paying the AI competence penalty? — https://leaddev.com/career-development/are-you-paying-the-ai-competence-penalty - Tech layoffs are making things worse for women in leadership — https://leaddev.com/career-development/tech-layoffs-are-making-things-worse-women-leadership - AI adoption has to be driven from the top — https://leaddev.com/ai/ai-adoption-has-to-be-driven-from-the-top ## 🗣️ Paraphrase 1. AI is new, but it's carrying old workplace inequality with it. 2. Research shows AI can raise productivity while widening existing gaps. 3. Women are 22% less likely than men to use AI regularly at work. 4. That's because they get less encouragement, less praise, and less training, not because they're less interested. 5. Good enablement meets each person where they are; treating everyone the same only helps people who'd adopt anyway. 6. Women who do use AI are judged more harshly: rated less competent than men who use it, and much less competent by men who don't use it. 7. That creates a loop — men get credit for using AI and build confidence; women get less and hold back. 8. Since AI skill is becoming a stand-in for leadership potential, a usage gap can become a promotion gap. 9. The fix is to remove the penalty for using AI, not just push adoption. 10. Measure your own gap instead of assuming your team is different. ## 📰 Newsletter **Subject: The AI penalty isn't equal** If your team is rolling out AI tools, read [this LeadDev piece](https://leaddev.com/ai/ai-widens-the-gender-gap) before your next adoption dashboard review. The numbers are well sourced. Per [Lean In](https://leanin.org/articles/findings/ai-women-gender-gap-data/), women are 22% less likely to use AI regularly at work — and men are 27% more likely to say their manager praised them for using it. Per Deloitte, 91% of men have had formal AI training versus 72% of women. The sharpest finding is from [HBR's study of 1,000+ engineers](https://hbr.org/2025/08/research-the-hidden-penalty-of-using-ai-at-work): identical code, rated less competent when a woman used AI to write it (13% penalty) than when a man did (6%). Among men who don't use AI themselves, the penalty for women hit 26%. Put that next to LeadDev's other piece this week, [on developers refusing AI](https://leaddev.com/ai/meet-the-developers-rejecting-ai) — four men who can afford to walk away. Both pieces point to the same conclusion: the costs of using AI, and of not using it, don't fall evenly. WordPress angle: our contributor spaces are increasingly AI-assisted. If women contributors get tougher code review when they use AI, we'd be repeating this in the open. Measuring it in review data is doable. One editing note: the closing advice ("measure the exact gap") is attributed to "Decrease" — probably DeCesare — right after a paragraph about Haines. Check the live page before quoting it. ## ✅ Action Items **DevRel** - When teaching AI tools, design enablement that meets people where they are (Haines's point) rather than one-size workshops. - Make AI use visible and praised evenly in community showcases. **OSPO** - Code review in open source is where the competence penalty would show up. Consider checking whether review outcomes differ by contributor gender and AI disclosure. **Community** - Track gender balance in AI-related events and contributor programs; the data suggests unequal encouragement is the main driver. **WordPress subteams** - *Core AI* — the most direct link: if AI-assisted contributions are reviewed more harshly depending on who submits them, the team should know. A disclosure-and-review audit would show it. - *Community* — mentorship and encouragement for AI experimentation should reach women contributors deliberately, since the gap comes from encouragement. - *Training* — Learn's AI content should be adaptive (skeptic to daily user) and should address the penalty openly. - *Core* / *Plugins* — reviewer guidelines could state that AI assistance isn't a competence signal either way. - *Marketing* — represent women using AI in WordPress showcases and case studies. - *Core Performance*, *Core Program Team*, *Mobile*, *Playground*, *Support*, *Design*, *Documentation*, *Photos*, *TV*, *Polyglots*, *Accessibility*, *Meta*, *CLI*, *Hosting*, *Tide*, *Openverse*, *Themes* — no direct action. **The WPOCC** - A community collective could publish norms for AI disclosure that avoid penalizing contributors, and could collect gap data across projects. **FAIR** - Not applicable to this source directly; no package-distribution content. ### → Things 3 - [ ] Read the HBR competence-penalty study's method before citing the 13%/6%/26% figures - [ ] Consider whether WordPress contributor review data could test the penalty ## 🧪 Hypotheses 1. **Encouragement closes the gap.** Teams where managers explicitly encourage and praise AI use equally will show a smaller gender gap in adoption. Testable with a manager intervention. 2. **The penalty shows up in real review data.** In code review, pull requests disclosed as AI-assisted will receive more change requests when submitted by women. Testable in open-source repositories with disclosure labels. 3. **Non-users penalize most.** Reviewers who don't use AI will penalize AI-assisted work more, especially by women (as HBR found). Testable in replication. 4. **The usage gap predicts promotion outcomes.** Within a year, AI usage will correlate with high-profile project assignments, and gender will mediate that. Testable with HR data. 5. **Adaptive enablement outperforms uniform training.** Tiered programs will raise adoption among late adopters more than uniform workshops. Testable with an A/B rollout. ## 📊 Methodology **Design:** an explainer built on four secondary sources plus two expert interviews. No original data. **Evidence quality:** - HBR experimental study (1,000+ engineers) — the strongest: an experimental design with revealed gender. The article doesn't give the study's country, sampling, or effect sizes' confidence intervals. - Lean In survey (1,000+ Americans) — descriptive, US-only; the "22% less likely" is a relative difference whose base rates aren't given. - Deloitte figures — percentages without sample details in the article. - HBS paper — cited for a general conclusion; its specific numbers aren't reported. - Expert quotes (DeCesare, Haines) — interpretation, not evidence. **Gaps:** no measure of skill, so the takeaway "not a skills gap" is inferred rather than shown; relative percentages without base rates; no data on non-binary people; one misattributed quote. ## ❓ Questions - What are the base rates behind "22% less likely"? A 22% relative gap means very different things at 20% versus 60% usage. - Did any cited study measure AI skill? If not, "not a skills gap" is an inference. - Who said the closing line — DeCesare or Haines? - Does the HBR penalty replicate outside the study population and country? - Would anonymous or blind code review remove the penalty? - How does the refusal trend in LeadDev's other piece interact with this — do women refuse AI for different reasons? ## 🔮 Future Research - Replicate the HBR competence-penalty design in open-source code review with real disclosure labels. - Longitudinal data linking AI usage, gender, and promotion outcomes. - Studies including non-binary and intersectional identities, which the cited research doesn't cover. - Intervention trials on manager encouragement and praise. - Comparing the gap across countries, given the Lean In data is US-only. ## 💡 Implications **Practice.** Managers should measure AI usage and outcomes by group, praise usage evenly, and tier enablement by readiness. Reviewers should be told explicitly that AI assistance isn't a competence signal. **Policy.** If AI fluency drives promotion, unequal access to encouragement and training becomes an equal-opportunity issue. HR policies on AI should cover evaluation, not just access. **Security.** Not applicable to this source in a technical sense. The nearest link: if people hide AI use to avoid a penalty, organizations lose visibility into which code and content was AI-generated, which does have security and quality consequences. **Education.** Training programs need to address confidence and the social risk of experimenting, not just tool skills. **Funding.** Research funders should support replication and intersectional work; the evidence base is small. ## 🧠 Learning Styles Assessment - **Visual** — weak: a hero image; no charts, though the numbers invite one. - **Aural** — not applicable; text only. - **Verbal** — strong: clear statistics with interpretive quotes. - **Kinesthetic** — moderate: ends with an action (measure your gap), but no method is given. - **Logical** — strong: a clear causal chain from encouragement to usage to evaluation to career outcome. - **Social** — moderate: expert voices, framed around team dynamics. - **Solitary** — moderate: invites reflection on your own team. - **Naturalistic** — not applicable to this source; no natural-world or classification content. **Reading level:** about grade 10. Short and statistic-dense; readers need to track four sources and several percentages. ## 📊 Accessibility Assessment Mapped to WCAG 2.2 AA, from the Reader capture. 1. **Text alternatives (1.1.1)** — the hero image's alt text is "Article hero fill the gaps-03," a filename, not a description. An ad image's alt ("NYC 27 Pre sale ticket block image") appears mid-content. *Fail.* 2. **Structure (1.3.1)** — clear headings and a key-takeaways box. *Pass.* 3. **Link purpose (2.4.4)** — "AI adoption" is used twice for the same link, and "using AI" is vague; most research links are descriptive. *Partial.* 4. **Readable language (3.1.5)** — plain language. *Pass.* 5. **Contrast (1.4.3)** — not assessable. 6. **Data presentation** — statistics are in prose only; a table would help screen-reader users compare figures. *Partial.* 7. **Consistent identification (3.2.4)** — not assessable from the capture. **Top fix:** replace the filename alt text with a description, and give the ad image empty alt. ## 🌍 DEIB Assessment **Cultural breadth.** US-centric: Lean In's data is American and the framing is US corporate. The HBS paper is described as "global," but the article doesn't report non-US findings. **Inclusive language.** Respectful; it locates the problem in conditions and evaluation, not in women. **Neurodiversity.** Not addressed. Relevant: social-risk calculations ("what does it cost me if this goes wrong in front of the wrong person?") also weigh on neurodivergent employees. **Representation.** Binary gender only; non-binary and trans experiences are absent, as are race and other intersecting identities. Two women experts are quoted, which is appropriate to the subject. **Psychological safety.** This is the article's real subject: whether people feel permitted to experiment. It names psychological safety as the variable without using the term. ## ✅ Logical Validity ### Strong reasoning - Four independent sources point in the same direction, and all are linked. - It separates access (encouragement, training) from evaluation (the competence penalty), which are different problems with different fixes. - The HBR finding comes from an experiment with identical code, which isolates the bias well. - The recommended fix (measure your own gap) is falsifiable. ### Fallacies 1. **Unsupported negative claim.** *Where:* takeaways — "not a skills gap." *Why it fails:* no cited study measures skill; the evidence shows encouragement and evaluation differences, which is consistent with no skills gap but doesn't prove it. *Fix:* "driven by encouragement and training differences." 2. **Claim drift.** *Where:* body "22% less likely… to use AI regularly" becomes takeaways "use AI at work 22% less." *Why it fails:* likelihood of regular use isn't amount of use. *Fix:* use the body's wording. 3. **Relative statistics without base rates.** *Where:* "22% less likely," "27% more likely." *Why it matters:* readers can't judge magnitude. *Fix:* give absolute percentages. 4. **Misattribution.** *Where:* "Decrease says." *Why it matters:* a quote attributed to a nonexistent name right after another expert's paragraph. *Fix:* correct the name. 5. **Speculative mechanism presented as description.** *Where:* DeCesare's feedback loop ("His brain registers…"). *Why it matters:* plausible but unmeasured. *Fix:* frame as a hypothesis. ### Tone Measured and evidence-led, with a clear advocacy goal. It avoids alarmism and blame. ## 🛡️ DARVO Analysis **Does the source perform DARVO?** No. **Is DARVO described?** Not directly. The competence penalty has a related shape: when a woman uses AI and gets penalized for it, the person adopting a tool the organization promotes is treated as the less competent party. That's a structural bias rather than a DARVO sequence (no denial, attack, or claimed victimhood is described). **Assessment:** not applicable to this source beyond noting that absence. ## 🎯 Conclusion **Bottom line.** A short, well-sourced piece showing that AI adoption is uneven by gender for two reasons: women get less encouragement and training to use it, and are judged more harshly when they do (a 13% competence penalty against 6% for men, per HBR). The fix it proposes — measure your own gap and remove the penalty — is concrete. Its weak points are editorial: a takeaways box that overstates the body, relative statistics without base rates, and a garbled attribution on the final quote. **Next steps.** 1. Read the HBR study before quoting its percentages. 2. Pair this with LeadDev's piece on developers refusing AI: using and refusing AI both carry career penalties, and neither falls evenly. 3. Consider a code-review audit in a WordPress contributor space to see whether the penalty appears there. ## 🧱 Refactor Appendix ### Literature Split Pad **L1 — Adoption gap.** Women are 22% less likely than men to use AI regularly at work (Lean In, 1,000+ Americans). Source: LeadDev, 2026-09-21. **L2 — Unequal conditions.** Encouragement 93% vs 84%; formal training 91% vs 72% (Deloitte); men 27% more likely to be praised (Lean In). **L3 — Competence penalty.** AI-assisted code rated 9% worse; women 13% vs men 6%; 26% from male non-users (HBR, 1,000+ engineers). **L4 — Feedback loop.** Encouragement and praise build a user identity for men; less credit keeps women from building it (DeCesare). **L5 — Usage gap becomes career gap.** AI fluency is becoming a proxy for leadership potential. #### Atomic Split Pad **A1 — Adoption gaps come from permission, not ability.** Who gets encouraged, trained, and praised predicts who adopts. Links: [[Psychological safety]], [[Enablement design]]. **A2 — The same tool, used by different people, is judged differently.** The competence penalty is steeper for women. Links: [[AI competence penalty]], [[Code review bias]]. **A3 — When a skill becomes a proxy for leadership, access to it becomes a career issue.** Links: [[Promotion criteria]]. **A4 — Uniform enablement reaches only early adopters.** Meet people where they are. Links: [[Tiered training]]. **A5 — Relative statistics need base rates.** "22% less likely" means little without the starting number. Links: [[Statistical literacy]]. #### Fleeting Queue - Author byline not captured; confirm on the live page. - Resolve "Decrease says" — DeCesare or Haines? - Get absolute usage rates from the Lean In data. - Check whether HBR's study covered non-US engineers. <!-- deep-read:sections end --> 📄 **Full text:** [[Full Document Contents/Articles/AI Widens the Gender Gap.md|AI Widens the Gender Gap by LeadDev]] ## Highlights - Engineers rated [AI-assisted code](https://leaddev.com/ai/ai-generated-code-sparks-production-confidence-crisis) 9% worse than identical code they believed engineers had written alone. When researchers revealed the coder’s gender, respondents rated female engineers who used AI 13% less competent, compared with 6% for male engineers. ([View Highlight](https://read.readwise.io/read/01m31yr0ed3hrm2b0j72jk5cr1)) - Note: Atom A2 — the same tool, used by different people, is judged differently. HBR experimental study, 1,000+ engineers. The strongest evidence in the piece. - Tags: gender-equity, atom/atomic/h1, workplace-bias - “Start by measuring the exact gap instead of assuming your culture is the exception,” Decrease says. ([View Highlight](https://read.readwise.io/read/01m31yr32rq0zas4pg9jmcyc3z)) - Note: The article's recommendation, plus an attribution defect: "Decrease" is likely a typo for DeCesare, and the quote follows a paragraph about Haines. Verify the speaker before quoting. - Tags: fallacy/misattribution, atom/literature/h2, engineering-management