Article
Making AI work for everyone
Women report less AI training, mentorship and encouragement at work, and that support helps everyone who gets it. What People teams can do in 90 days.
Most AI rollouts assume that once everyone has a license, everyone has the same chance. They don’t. Access to the tool is the easy part. Access to the help that makes the tool useful is uneven, and the gaps follow lines HR already knows how to see.
New research from Harvard’s Shorenstein Center puts numbers on one of those gaps. It’s worth reading in full. Here is what it means for a People team planning next year.
The gap starts with support
Among employed adults, women are less likely than men to say they get AI training, mentorship, or encouragement to use AI at work. Training shows the widest gap: 64% of women report no AI training at work, against 49% of men (Shorenstein Center, 2026).
Use follows the same line. 46% of women say they never use AI at work, compared with 36% of men (Shorenstein Center, 2026).
It’s tempting to read the second number as a choice. The first number tells a different story. Many people who don’t use AI at work were never shown how, never had a colleague walk them through it, and never heard a manager say it was expected. That isn’t only a motivation problem. It’s a distribution problem, and distribution is something an organization controls.
Gender is not the only line, either. Only 26.3% of AI users say everyone in their organization benefits equally from AI (Charter AI Pulse, 2026). Most people can see that the gains are landing unevenly. Few organizations are measuring where.
Support works for everyone
The more useful finding is what happens when the support is there. Training, mentorship, and encouragement at work are associated with more use, more confidence, and more perceived value, for women and men alike.
So this isn’t a side program for one group. It’s the core of the rollout. Structured support raises the floor for everyone who gets it, and the people who get the least today have the most to gain.
One honest caution from the research: gender gaps often persist even among people who do get training and mentorship. Support narrows the distance. It doesn’t close it on its own, which is why the steps below include measuring it.
The barriers are trust and privacy
Ask people why they don’t use AI more, and the top two answers for women and men are the same: privacy concerns and distrust of what AI produces. Fear of being judged for using AI ranks near the bottom, chosen by 5% of women and 6% of men (Shorenstein Center, 2026).
That changes what a good program says out loud. A pep talk about not being afraid misses the point. People want straight answers to practical questions:
- What happens to the data I put in?
- Which tools are approved, and for what?
- Who can see what I type?
- How do I know when the output is wrong?
If your program can’t answer those in plain language on day one, expect the people with the most reasonable doubts to sit it out. They aren’t resisting. They’re asking the right questions.
Uncertainty isn’t ignorance
In the same research, women answered “not sure” more often than men on factual questions about AI. It would be easy to read that as knowing less. The researchers suggest part of the gap may be confidence rather than knowledge. Asked whether they recognized a list of people in AI that included two invented names, men were more likely to say they recognized every name, the invented ones included.
That matters for how you teach. A good fluency program doesn’t reward the loudest answer. It builds calibrated confidence: knowing what you know, saying what you don’t, and checking before you commit. That is exactly the habit you want once people start acting on AI output. “Not sure, let me check” is a skill. Bravado is a risk. My verification checklist turns that habit into a routine.
What People teams can do in 90 days
None of this needs a new program. It needs the rollout you already have to be deliberate about who gets help.
- Audit who gets access. Pull the data you already have: who was invited to AI training, who attended, who has a mentor or a go-to colleague for AI, which teams have a manager who uses it where people can see. Break it down by the groups you already track for other programs. Look at the gaps before the averages.
- Make support structural, not informal. Informal help flows to people who already have networks. Replace “ask around” with cohorts people are enrolled in, mentors who are assigned, and practice time on the calendar. If getting help depends on knowing who to ask, it will be uneven.
- Answer privacy and data questions head-on. Write a one-page answer to the questions above with your Legal and Security partners. Put it in the first session, not an appendix.
- Ask managers to say it out loud. Encouragement is the cheapest support to fix. Ask every manager to tell their team, in a meeting, how they expect AI to be used and who to go to for help. Silence reads as “not for you.”
- Track access by group, not just adoption. Add training, mentorship, and confidence to the measures you report, broken down the same way as step one. If the overall numbers rise while the gaps stay put, the rollout isn’t finished.
The point
Higher AI use isn’t the goal in itself, and the researchers say so plainly. The goal is that everyone has a fair chance to decide when AI is useful in their work, and the support to use it well. Making sure that support reaches everyone is a People team’s job. The data says most organizations haven’t done it yet.
The readiness check takes five minutes and shows where your People team stands across six dimensions, and where the next moves are.