Evidence on how Nepal adopts AI
Most claims about AI adoption in Nepal are assertions. I run the surveys that turn them into measured findings, and publish the results so policymakers, employers and vendors can act on numbers rather than impressions.
The Adoption Paradox
How Kathmandu's professionals use AI faster than they understand it, and faster than anyone is governing it
A cross-sector study of AI awareness, adoption, trust and governance readiness across eight sectors in Kathmandu, based on 334 face to face interviews with working professionals.
Generative AI reached Nepal through mobile phones rather than through corporate procurement, IT budgets or government programmes. That single fact shapes everything else in the report: adoption looks high, but capability, workplace integration and governance did not follow behind it. The study measures those layers separately instead of assuming they move together.
The result is a market where the usual signals mislead. Awareness is effectively universal and personal use is high, but the depth of that use is thin and unevenly distributed, and the two channels through which a technology is normally governed, employer training and national policy, barely register.
Awareness is complete; competence is not
98% have heard of AI and 81% use it personally, but only 39% say they understand it well enough to use it regularly, and just 22% combine that understanding with daily use. The funnel loses 76 points between hearing about AI and using it capably.
AI arrived through the phone, not the workplace
87% first learned about AI through social media. Only 8.5% learned through workplace or professional training and 10.7% through formal education. On the most consequential general-purpose technology of the decade, Nepal's institutions reached fewer than one respondent in ten.
The toolset is one product deep
ChatGPT is known to 91% of respondents; the next best known tool reaches 56% and no other exceeds 26%. AI capability in Kathmandu today effectively means general-purpose chatbot literacy, which is a narrow foundation for sector-specific value.
There is a generational cliff, not a gradient
Daily use runs at 52% among 20 to 29 year olds and falls to 24% in both the 30s and 40s bands. Once age is controlled for, education level explains almost nothing.
Trust runs ahead of understanding
58% trust AI for important decisions, either completely or with human oversight, while only 39% claim to understand it well. Trust is highest where scrutiny should be greatest: 86% net trust in media and journalism against 37% in healthcare.
The workforce is anxious but not acting
61% are concerned about AI affecting job security in their sector, yet only 40% are actively learning new skills. In healthcare, 63% are concerned and 20% are acting.
The governance gap is the headline risk
92% want AI regulated and 78% say government should lead, but only 46% know Nepal has published a draft national AI policy: a 45-point gap between regulatory demand and awareness of the regulatory response.
The immediate problem is distribution, not drafting. Nepali-language summaries, sector guidance for healthcare and legal practice, and dissemination through the same social channels that carried AI into the country.
Convert shadow use into supervised use. The first move is not a training budget, it is an acceptable-use policy and a register of what staff are already doing.
Sell depth, not access. The market does not need to be told AI exists; it needs to move from general-purpose chatbot use to sector-specific application.
334 valid face to face interviews from 343 contacts, conducted in Nepali or English at the respondent's preference, March to April 2025. A purposive sample designed for coverage of eight sectors, which supports description and comparison between groups but not projection to population totals. Question bases vary between n=280 and n=334 and are printed with every exhibit. Reviewed before publication by an independent editorial board.
The study measures what respondents report doing, not what AI delivered: no figure in it is a productivity estimate. Understanding, trust and preparedness are self-reported and untested, every relationship is an association measured at one moment, and the findings describe 334 professionals in Kathmandu district rather than Nepal as a whole.
Research, briefings and policy notes
I run original survey research on AI adoption and readiness, and write it up for the people who have to act on it. If you need evidence for a policy, a strategy or a board paper, tell me the decision you are trying to make.
Discuss a study