AI-generated article. This article was researched and drafted using AI tools and published automatically, and its featured image was generated by AI. Facts are drawn from the sources cited in the text.
Two women run comparable businesses. Similar size, similar sector, similar ambition. One works from a home office with fixed broadband and a door that closes. The other shares a phone, travels to reach a reliable signal, and cannot always control the money her business earns. Handing both the same AI tool and the same free training does not produce the same result, and the reason has very little to do with either woman. This is the part of the global AI gender gap that gets flattened whenever the story is told as a skills problem.
The Cherie Blair Foundation for Women, working with Intuit and the World Bank’s Women, Business and the Law project, surveyed more than 3,000 women business owners across 66 countries in low and middle income economies. The headline finding was enthusiasm. Around 82% already use AI in some form. The complication is where that use stops, a pattern explored in more depth in the difference between adopting a tool and embedding it.
Eighteen hours before the business day starts
The same research put average time spent on caregiving at around 18 hours a week, carried alongside running a company. The Foundation’s own framing of the result is unusually direct: what holds women entrepreneurs back is not skill and not confidence. It is time poverty.
That distinction changes what a sensible intervention looks like. A six-week evening course assumes free evenings. A webinar assumes a quiet room. A tool that requires an uninterrupted hour to configure assumes an uninterrupted hour exists. Most AI training is designed, without anyone deciding this, for a person whose time is their own.
What widens the global AI gender gap
Geography adds constraints that rarely make it into adoption statistics. Research on AI deployment in the Global South, including work published by New America, describes structural barriers around mobility, time use, safety and financial autonomy that limit participation even in programs explicitly designed for inclusion. A training session in a location a woman cannot safely travel to in the evening is not a training session she declined.
UNESCO launched an outlook study on artificial intelligence and gender in South Asia, and researchers writing for Brookings have argued for feminist and Global South perspectives on AI-supported learning environments specifically, on the grounds that tools built around one set of assumptions travel badly. The pattern repeats across the literature: barriers compound for people already excluded, so rural, remote and less formally educated women meet the gap in a sharper form.
None of these are confidence issues. They are constraints on time, movement, connectivity and control over money. An operational constraint and a control constraint, sitting underneath what gets reported as low adoption.
What this means outside those 66 countries
It would be comfortable to read this as a story about somewhere else. The underlying design flaw is not regional. Any sole trader juggling care work meets a version of the same problem, which is why learning tends to work better in small doses than in dedicated blocks. The Global South evidence simply makes the constraint impossible to mistake for a preference.
There is a real opportunity in this too, and the numbers support saying so. Adoption at 82% is not a population that needs persuading. The enthusiasm is already there, which is a much better starting position than indifference. The gap is between wanting to use these tools well and having the conditions to do it.
Something that actually fits
Research on what helps keeps returning to peer learning. Spaces where business owners work through implementation problems with others facing similar constraints tend to outperform formal curricula, partly because they run on shorter, more flexible units of attention and partly because the advice arrives already adapted.
For anyone designing training, the useful test is simple. Could this be completed in fifteen minute pieces, on a phone, interrupted, without losing the thread? If not, the program will select for women who already had the least trouble, and then report their success as evidence it worked.
For a woman running a business inside these constraints, the honest version is that fifteen minutes on one repeated task is not a compromise. Given what the research says about where time actually goes, it is the realistic unit. Starting there puts anyone ahead of waiting for a free afternoon that is not coming.
