From scepticism to action: practical AI recipes for under-resourced cultural teams

Most cultural organisations I speak to about AI say some version of the same thing: "We know we should be doing something, but we don't know where to start." You've probably got an AI policy by now (if not CultureHive has published a good one). What you might not have is a practical next step. Something you could try on a Tuesday afternoon that doesn't require a board paper or a data scientist.

That's a reasonable position. The technology moves fast, the hype is exhausting, and the last thing an under-resourced arts team needs is another consultant telling them to "embrace transformation." You've heard it before. You've been burned before. Digital strategies that went nowhere. CRM migrations that cost twice the budget and delivered half the promise. Now AI shows up and says trust me.

I get the scepticism. It's mostly deserved from some of the behaviour of OpenAI et al. But I also think there's a gap between healthy scepticism and productive action that's leaving a lot of cultural organisations stuck.

That's what we tried to build.

What the recipes are

AI Recipes for Charities  is a free collection of 89 practical guides - we call them recipes - each built around a specific problem that charity and cultural organisations face. Not "implement AI across the organisation." More like: "You've got a Heritage Fund bid due on Friday and nobody's had time to review it properly. Here's how to use AI to stress-test it before you submit."

Each recipe has a difficulty level (beginner, intermediate, advanced), a category (data analysis, fundraising, communications, operations, service delivery, impact measurement, compliance), and tells you what you start with, what you end up with, and what it costs. Most of the beginner recipes cost nothing, just a free ChatGPT or Claude account and twenty minutes.

The recipes sit alongside the AI Playbook for Charities, a longer resource with case studies and context. The playbook explains the thinking. The recipes are what you do on Monday morning. Both are free.

Why we built them

My name’s Edd Baldry. I run Make Sense Of It, a small applied AI agency, with my co-founder Suzanne Begley. We work mostly with charities and cultural organisations - Breast Cancer Now, the National Lottery Heritage Fund, NHS trust spin-outs. With the Arts Marketing Association we also built Goose, an AI tool specifically for the heritage sector, which I'll come back to.

The short version of why the recipes exist: I've spent twenty years in and around charities, and I've watched the sector miss two major technology shifts - the web and social media - for entirely understandable structural reasons. People were busy, underfunded, and the technology didn't fit into any framework they had for making decisions. AI is the third shift, and the structural barriers haven't changed. The recipes are an attempt to lower the starting point. Instead of asking people to understand the technology first and then figure out what to do with it, we start with the problem and work backwards.

Recipes for cultural organisations

Here are four that are directly relevant to arts, heritage, and cultural work. Each takes less than an hour.

Stress-test an Arts Council or Heritage Fund bid (Beginner - Fundraising)

Paste your draft bid into ChatGPT or Claude and ask it to play the role of a sceptical panel member. Ask it to find weak evidence, vague outcomes, and anything that doesn't answer the question being asked. You're not getting AI to write the bid - what you know about the work and the community is the thing that matters. You're getting a second reader when nobody else on the team has time. One charity I worked with has this embedded in their process now because AI consistently did better at spotting gaps than their internal reviews.

Turn visitor data into a narrative for funders (Beginner - Impact Measurement)

You've got the numbers, whether they’re attendance figures, survey scores, participation demographics. You need 500 words that turn them into a story for your next board report or funding return. Give the data to an AI tool with clear instructions about tone and audience and you'll have a workable first draft in minutes instead of half a day. You'll need to edit it given that the AI doesn't know your programme or your community as well as you do. But turning raw data into the first-draft of structured prose is what these tools were originally built to do and they’re still good at it.

Find patterns in post-show or visitor feedback (Intermediate - Data Analysis)

If you've got more than a hundred free-text responses, like post-show feedback, visitor comments, community consultation returns, AI can surface themes you'd miss reading them one by one. Upload the responses and ask for a thematic analysis. What keeps coming up? What complaints cluster? Where are people saying the same thing in different words? This is particularly useful after a season ends or a touring show closes, when the volume of feedback makes manual analysis impractical and the insights would be most useful for programming decisions. If you're using the free versions of Claude, Gemini or ChatGPT, be aware that your conversations may be used to train future versions of these tools. Before pasting visitor feedback or community responses, remove any personal details like names, email addresses, or identifying information. Most of your feedback is already anonymous. This is an important precaution to make sure nothing identifying gets into the training data. But I bet you’re sitting on a mountain of already anonymous data that you could use!

Brief yourself before a board meeting (Beginner - Operations)

Feed the agenda papers into an AI tool and ask for a plain-English summary of each item, the decision being asked for, and the risks. You still need to read the papers. But having a map before you start changes how you read them, especially for trustees on multiple boards who are fitting preparation into evenings and weekends.

What we learned building Goose

The recipes are one side of this. The other is Goose - a production AI tool we built for the Arts Marketing Association and funded by the National Lottery Heritage Fund. The tool is designed specifically for heritage professionals in marketing.

The heritage sector has a particular version of the resource problem. A marketing team might be one person. That person is also doing community engagement, managing volunteers, writing funding bids, and updating the website. They don't need a chatbot. They need colleagues - someone to pressure-test a campaign strategy, give a fresh perspective on audience development, or help think through how to position a Grade II listed building that's open four days a week.

Goose provides 21 specialist thinking partners - a community engagement manager, a fundraising specialist, a heritage interpretation expert, a digital marketing strategist, among others. Users bring multiple partners into a conversation and get different perspectives, grounded in heritage-sector knowledge. They can also create their own. It's been running in production since August 2025, used daily by heritage professionals across the country.

Building Goose taught us something that shaped the recipes. Heritage professionals don't want AI to do their job. They want it to make the job feel less lonely. The most-used feature isn't content generation - it's the ability to get a second opinion on a strategy or a draft when there's nobody else in the building to ask. That's augmentation, not automation. As one of the co-creators said:

Goose is a thinking partner, not a replacement for thinking.

Augmentation vs delegation

That distinction - augmentation vs delegation - is the most useful thing to come out of writing the recipes.

Some tasks should be augmented by AI: the tool challenges your thinking and gives you a starting point to react to. Other tasks can be delegated - reformatting data, cleaning up meeting notes, translating text.

The test is simple. Would handing this task entirely to AI make the work worse because human judgment is the point? If you're writing interpretation panels for an exhibition, your curatorial knowledge and understanding of your visitors is the work - AI can help you draft, but the thinking has to be yours. If you're converting a spreadsheet of loan agreements into a summary table, let the machine do it.

That's a trust distinction, not a technical one. It gives you a way to decide where AI fits in your organisation without handing over the things that make your work yours.

What doesn't work

Here's where it gets messier .

AI confidently makes things up. It'll invent statistics, fabricate quotes, produce plausible-sounding claims about your organisation that aren't true. Anything factual that comes out of an AI tool needs checking. If you take one thing from this article, make it that.

The recipes that involve generating text - fundraising copy, social media posts, press releases - are the weakest category. The output is usable as a starting point but it reads like what it is: a statistically average version of everything the model has seen. If your organisation has a distinctive voice, AI will flatten it. There are glimpses of tools - like Claude Cowork - where this doesn’t happen but they’re still for early adopters at the moment. For now, the safe approach is to understand that you’ll need to rewrite the draft that comes out of AI.

Where to start

Visit airecipesforcharities.com and pick one recipe that matches a problem you actually have this week. Not the most ambitious one. The one where you think: yes, that's annoying, and I'd like it to be easier. Try it. See what happens.

The recipes are free and will stay free. We built them because we'd spent twenty years watching the sector get left behind by technology shifts and figured the least we could do was make the starting point easier.

Edd Baldry, Make Sense Of It

If you want to talk about what AI could look like for your specific organisation - whether that's a tool like Goose or something much simpler - we're at Make Sense Of It. But start with a recipe. You'll learn more from twenty minutes of trying than from any amount of reading about it.