For the past few years, I’ve watched various consultants, freelance writers and fundraisers dipping their toes into ChatGPT to write fundraising copy. They wonder if this tool will do the work better, faster, more efficiently – or if it can even do it at all.
So, they use a quick-and-dirty prompt – “Write me a two-page fundraising letter for my year-end campaign” – in the world’s most widely-used, generalised, AI tool. And when junk copy comes back, devoid of fundraising best-practice and emotion, I see the same conclusion drawn time and time again: “AI isn’t a good fundraising writer.”
But getting one bad letter from an AI model, and concluding that AI can’t write for fundraising, kind of misses the mark. And, it risks completely overlooking the real potential that AI has to operate as a tool in our fundraising toolbox.
Let me put it another way: Think about teaching someone to cook. You wouldn’t expect someone who’s just learned how to crack an egg to have the technique to whip up a flawless French omelette, right? And you wouldn’t tell them they’ll never become a proficient chef when they burn a dish. Just because they can’t do one thing perfectly YET doesn’t mean they can’t EVER – or that they can’t do other things, too. They need you to teach them, first.
The Books3 scandal
Making that French omelette starts with a recipe and ingredients. Making AI started with an algorithm and data.
Between 2020-2023, companies like OpenAI and Meta used a dataset called Books3, which was around 190,000 pirated books taken from a shadow library website, to train their LLMs. They didn’t ask, they didn’t pay, and if you have a subscription to the Atlantic you can read Margaret Atwood’s take (TLDR: not cool, not good, not okay, and it’s a crap poet).
Basically, the goal was to see if they could get AI to write copy that sounded like a human wrote it. You know the results of this, because it produced the LLMs that you and I use today.
Because yes, this technology could write like a person. And not only that, it could also re-create the patterns in a particular work, and could produce something that echoes a selected author’s distinctive literary voice— things like rhythm, cadence, and vocabulary.
It didn’t just learn to write like a person. It could learn to specialize its voice to sound like a specific person.
Which made us think: if AI can be trained to mimic an author’s particular literary style, like that of Zadie Smith or Stephen King, couldn’t it also be trained (and should we train it) to write in a particular donor’s voice using the best practices of effective direct mail copy?
Maybe AI wasn’t a good copywriter… yet.
Creating glitter: Our AI fundraising copy collaborator
That question led us to develop Glitter, our own specialized AI tool built specifically for fundraising copywriting, at Good Works.
Over 18 months of development, we’ve taught Glitter to understand what makes fundraising copy effective—the patterns that drive response, the emotional triggers that inspire giving, the technical requirements that improve readability, and so much more.
The results have been impressive.
Since working with Glitter, we’re drafting tighter copy that better meets our agency’s standards, and does so more consistently.
In part, this is also because Glitter demands better directive inputs from our team. This means more clear and concrete articulation of the strategy, more detailed direction on which way to tell a story, a solid interview and carefully-cultivated background materials, and specific instructions on which copywriting best practices to draw upon.
Just think about the fiascos of Amelia Bedelia, the much-beloved fictional character from my childhood, that occurred because she’s very literal. What happens when she’s given the instruction to “string the beans”? She hangs them outside on a clothesline, but does not remove the hard stringy part from each one. When she’s asked to dress the chicken, she carefully picks out fancy overalls and socks to adorn the poultry, yet doesn’t apply an ounce of seasoning.
Working with AI to draft direct mail copy is a very human-driven process. You can’t trust that it will read between the lines and accurately interpret what you want it to do. You have to tell it very specifically what and how you want. Glitter demands that you be more thoughtful; otherwise, it generates crap copy.
We’re also finding that copy is also getting fewer revisions from both our team and clients. That’s because Glitter remembers the idiosyncrasies of a particular charity that humans are inclined to forget. Things like particular spelling preferences for “healthcare” versus “health care” in copy are no longer missed.
And, most importantly, Glitter also remembers the best-practice concepts that need to appear in every great letter. By this, I mean the things that humans are more inclined to forget in their own process of writing. These are the technical elements of fundraising copywriting upon which there has been no formal or consistent training across the sector. Direct mail writing is both a skill and a craft.
So now that we’ve taught Glitter how to construct the most perfect French omelette, what else can it cook? Because getting AI to write a single direct mail letter was never our end goal.
Beyond better writing: Personalization at scale
The real question isn’t, “Can AI write fundraising copy?” The real question is, “How can we use AI to do better fundraising?”
To us, the true power of AI in fundraising isn’t in writing one perfect appeal. It’s in creating personalized variations that speak directly to different donor identities and motivations—something that’s been financially untenable for most charities (although the for-profit sector has been doing it so long that they’ve reset consumer expectations, leading just about everyone to anticipate personalized marketing and brand experiences).
Personalization in annual giving is currently limited en masse to zero-party data that is entirely transactional: What amount was my gift, what date did I make my gift, and what appeal did I respond to? That’s what RFM (recency, frequency and monetary value), which most of our segmentation and KPIs are based on, tracks. A donor’s donation history.
We want to be able to use AI to democratize personalization in our fundraising work. We want small and mid-sized organizations to be able to create identity-based messaging at a scale previously available only to corporate giants with massive marketing budgets.
Consider a food bank sending appeals. Using Glitter, we can create variations that go beyond a couple variable paragraphs. We can craft entire letters grounded in the same theme, but which speak specifically to the different identities that different donors engage with through the connection with a cause:
- Compassion-focused: “Your kindness means a mother won’t have to see her children go to bed hungry tonight. Your compassion creates immediate relief for families facing impossible choices.”
- Fairness-focused: “Everyone deserves access to healthy food, regardless of economic circumstances. Your commitment to creating a more equitable community ensures resources reach those who need them most.”
- Respect-focused: “Your support provides dignified food assistance that honours each person’s humanity. When families visit our market-style pantry, they make their own choices about the food that’s right for their situation.”
This level of identity-based personalization – which has the potential to increase the funds raised by 20% or more! – is about creating a more meaningful connection between causes and their supporters. It’s about honouring donors’ unique relationships with causes they care about because it has the ability to speak to the identity of why they are when they give. And ultimately, it’s about raising more money for vital work that charities do to create a more just world.
The future is personal
The fundraising revolution won’t come from getting an out-of-the-box AI to write your direct mail appeal. It was never going to!
But it can come from creating a custom GPT, just like we did with Glitter. And that GPT will only be as good as you train it to be. For us, our lofty goal is to create personalized experiences for every donor—recognizing their unique identities, interests, and relationships with a given cause.
That’s the question we should be asking: How can AI help us speak to each donor as the individual they are?
Because this could change everything about how we connect with supporters.
A Footnote that Should Be the Headline: Glitter was an idea that we then built in collaboration with George Irish of Fundraising With AI. George was one of the OG digital fundraisers, so it’s not a surprise that he was at the forefront of AI developments in the charitable sector. For more than 18 months, we worked on this project together, in a total iterative fashion, wondering if we could teach AI to write like a fundraiser. And by golly, we did it! We launched Glitter with our clients in May 2025 and were just starting to talk about it publicly in conference presentations, articles and webinars. I can’t talk about Glitter without acknowledging George’s contribution. He died unexpectedly and suddenly in September in the great outdoors doing one of the things he loved best. You can learn more about George here.
Although Holly Wagg had an 8-track player in her childhood bedroom, she doesn’t consciously remember a time where you couldn’t fundraise without the internet. Holly has worked in the sector for more than 30 years, and is now Partner and CEO with Good Works, an integrated fundraising and communications agency that builds human connections between donors, causes and charities.
The views expressed in this article are the author’s alone and do not necessarily represent those of CharityVillage.com or any other individual or entity with whom the authors or website may be affiliated. CharityVillage.com is not liable for any content that may be considered offensive, inappropriate, defamatory, or inaccurate or in breach of third-party rights of privacy, copyright, or trademark.
