Why we updated our playbook for charities

When we published the first edition of the AI Playbook in March 2025, we thought it might have a shelf life of two years. It lasted about six months before the gaps became obvious.

The technology moved. That was expected. What we hadn't expected was how quickly the sector would move too. In early 2025 we were still having conversations with charity leaders who hadn't tried AI themselves. By late 2025, almost everyone had. The conversation shifted from "should we?" to "where does this actually help?" and the first edition didn't have enough answers to that second question.

From speculation to evidence

The 2025 edition was, honestly, pretty speculative. We were confident about the direction but we were working with fewer case studies than we'd have liked, and many of the examples we could point to used deterministic AI rather than the generative AI tools most charities were asking about. We were mapping territory that was still being explored.

The 2026 edition has a lot more evidence to back itself up with. About 50 case studies now, covering fundraising, service delivery, operations, impact measurement, communications, and data. Not theoretical examples. Charities doing actual work with AI and learning what happens when you do. That changes the texture of the advice. We can say "this works and here's who did it" rather than "this should work and here's why we think so."

We also have a much better understanding of where the edges are. The jagged frontier, the line between what AI handles well and where it falls over, is better mapped now than it was a year ago. More importantly, we understand better how generative AI fits within software pipelines to make it safe: where you need validation layers, where human review is essential, where you can let AI run and trust the output. The 2025 edition talked about AI capabilities. The 2026 edition talks about AI in context, as part of systems that actually work.

The sector context shifted too. Financial pressures intensified through 2025. Redundancies, funding cuts, increased demand. The case for AI helping stretched teams became stronger and more urgent, but so did the constraint that implementing anything requires investment that feels impossible right now. We tried to deal with that tension directly rather than pretending it doesn't exist.

Why bother

People ask us this, politely or not. A small agency writing a free playbook for an entire sector is not an obviously rational use of time. We could be billing those hours to clients.

The honest answer hasn't changed from last year. Writing the playbook makes us better at our work. The research required to produce something comprehensive forces us to understand things we might otherwise skim. It maps the territory, and mapping the territory shows us where the territory gets hard enough that charities need someone who can build within it. That's useful to us even if nobody else read it, which fortunately they do.

The less selfish answer: agencies working in this sector should share knowledge. The gap between what frontier AI labs are producing and what's practically useful for charities is wide, and it's not closing on its own. The language coming out of Silicon Valley is alienating. The hype cycle makes it harder, not easier, to make good decisions. If we can help charity leaders separate what's real from what's marketing, the whole sector benefits, including us.

Organised around how charities work, not how AI works

The 2025 edition was organised around AI concepts: what AI is, how it works, what it can do. The 2026 edition starts from the other end: fundraising, service delivery, operations, impact, communications, data. Each section includes practical guidance and links to specific AI Recipes for the use cases covered.

The 2025 edition came in at just over 11,000 words, already cut from an original 50,000. The 2026 edition is longer because there's more to cover, but we tried to keep the density high and the waffle low.

You can read the full playbook at aiplaybookforcharities.com/2026. If you read the first edition, the new one is worth the time. If you didn't, start with this one.