Introducing Nonprofits to Generative AI

As a product designer at Resilia, I was tasked alongside our PM and engineering lead with bringing generative AI to our nonprofit platform for the first time.

In prior discovery, we had repeatedly heard from our users — grassroots nonprofits — that they struggled to write donation appeals. The reality was that at small nonprofits, staff lacked the time and training to write copy that converted, which meant less money reaching the communities they serve.

These users were also new to AI. Almost none had used an LLM, so we saw two risks: that users wouldn’t know the limitations of AI, and that the AI itself would generate generic, non-mission aligned content.

“Write with AI,” our first feature, was designed to address these risks. I put it inside the rich text editor, where fundraisers were already writing, and added a short first-run explainer so trying it required no prior knowledge.

To keep output specific, every request combined the campaign’s own copy (autofilled, so no retyping), the nonprofit’s mission statement, and fundraising best practices from Resilia’s experts. Generated text stayed in its own dialog until a user chose to apply it.

In testing, 4 of 5 fundraisers rated it “very easy” to use and rated the output as “very good.” One user told us she’d been hesitant to use AI at all but would feel comfortable using this.

With that foundation, we took on a harder problem: improving Resilia Stories — our impact reporting tool for nonprofits. Nonprofits that don’t report regularly leave boards, donors, and funders in the dark. Nonprofits wanted to be better at reporting, but the blank first draft was where they got stuck the most.

So instead of a rewrite button, we brought AI in at template selection phase, where it could draft an entire story from scratch.

Relevant drafts depend on specific inputs. Alongside the nonprofit’s name and mission, I narrowed the form to four required fields, with goal and audience as select menus refined with subject matter experts, plus optional tone and length.

Our first usability test showed users wrote too little about their impact for the AI to work with. Adding examples to those fields greatly improved input in the second round of testing.

Generation could take 10+ seconds, so I designed a loading state that walked users through each step the AI was taking.

Rather than return plain text, the AI assembled the draft from Stories’ own text, image, and call-to-action blocks, producing a story users could edit like any other.