The Governance Gap
Read this first
This is an interpretation of research conducted by other people. We have run no survey and gathered no original data. Every figure below belongs to the study named beside it, and the argument connecting them is ours — which means it is arguable, and you are welcome to argue with it.
The numbers
A 2026 benchmark study of 346 nonprofits found that 92% now use AI in some capacity, while only 7% report major operational impact. Nearly half — 47% — have no AI governance policy at all, and 81% describe their use as ad hoc, with no documentation of what actually works.
A separate 2026 survey of 75 organisations reached the same place from a different direction: bandwidth, not budget, was the primary barrier, and unclear ownership of AI initiatives was the recurring point of failure.
What that combination means
Read together, these figures invert the usual pitch. The story the market tells is one of adoption — organisations that have not yet started, waiting to be shown the way in. The data describes something else entirely: organisations that started long ago, without a decision being made, and have no idea whether it is working.
A 92% adoption rate against a 7% impact rate is not an access problem. It is a problem of ownership, documentation and governance. And those three are conspicuously the things nobody sells, because they are unglamorous and they do not require a licence.
Why the gap persists
When no person is accountable for a piece of work, errors become nobody’s fault and quietly persist. When nothing is written down, the organisation cannot tell the difference between a practice that works and one that merely survived. And when there is no policy, staff make individually reasonable decisions — a free tier here, a pasted client record there — that no one would have approved as a policy.
None of that is solved by better tools, which is why better tools have not solved it.
What this does not say
It does not say AI addresses underfunding, understaffing, philanthropic inequity, or burnout rooted in inadequate staffing. Those are structural. The same 2026 data makes the distinction visible: the barriers named are governance and ownership, not capability and not cost. Any proposal that blurs the structural into the technical should be read as information about the vendor.
The specific relevance here
Pew Research data indicates Black teenagers use AI chatbots daily at higher rates — 35% — than Hispanic teenagers at 33% or white teenagers at 22%. High usage is not high literacy, and it is certainly not governance awareness or economic capture. Where usage runs ahead of understanding, the gap is not neutral: it is where value and control leave a community without anyone deciding that they should.
Jobs for the Future’s 2025 report Unlocking the Promise of AI for Black Learners and Workers recommends community-embedded training led by organisations deeply rooted in local context, and co-creation of tools reflecting lived experience. That is a named, independent endorsement of a delivery model, not of us.
Where this argument is weakest
The nonprofit benchmark study is a single sample of 346 organisations, self-selected into responding, and the definition of “major operational impact” is the study’s rather than an agreed measure. Two 2026 surveys agreeing with each other is a pattern worth taking seriously, not a settled finding — and we would rather say that here than have you discover it yourself.
Sources
- 2026 Nonprofit AI Adoption Report — Virtuous (benchmark study, 346 nonprofits)
- Unlocking the Promise of AI for Black Learners and Workers — Jobs for the Future
- Pew Research Center data on AI adoption among Black Americans, via secondary aggregation.