Claude Cowork for nonprofits: the mission gets the hours back.

The prompt mindset: role, task, context 0:00

Last time we compared the four major AI platforms - Claude, ChatGPT, Gemini, and Microsoft Copilot - what each one is good at, how they're priced for nonprofits, and how to figure out which fits your organization. Today is the sequel. You've picked your tool, or at least you're curious about one. Now the question is what you actually do with it.

AI is a very capable assistant, but it needs clear instructions. The formula I use is role, task, context. You tell AI what role to play, what you need it to do, and enough context to do it well.

Here is a bad prompt: "Write a grant proposal." That is like telling a new employee "do the new grants" and walking away. They don't know which grants you're talking about. There is no context.

Here is a good prompt: "Act as a nonprofit grant writer with 15 years of experience. Draft a two-paragraph statement of need for a workforce development program serving recently incarcerated adults in Wake County, North Carolina. The program has a 78% job placement rate and serves 120 participants annually." Same tool, completely different results.

There's one other thing I want you to internalize. You're the editor, not the author. AI gives you a strong first draft, but your job is to review, refine, and make it yours. That mindset shift alone takes some of the pressure off.

Grant writing 3:00

Grant writing is probably the single highest-value use case for AI in the nonprofit world. Most development staff spend 60 to 80% of their time on writing tasks: need statements, narratives, logic models, executive summaries. That is exactly where AI can shine.

There are about four specific places where it saves real time. Drafting a statement of need, where you paste in your numbers, your population, your geography, and it produces a funder-ready paragraph. Summarizing a full narrative into an executive summary. Adjusting tone to match a specific funder, because foundation language is a lot different from government language. And generating a logic model outline from a program description.

These aren't tricks. These are legitimate time savers that free you up to do the relationship work and the strategy work that AI can't do.

In the demo I paste in a real RFP requirement from a workforce development funder, give Claude the role and context it needs, and watch it produce a first-draft statement of need. Notice what I gave it: a role, a specific task, the funder's actual requirement language, and concrete program data. The more specific you are, the better the output.

What you just saw took 20 to 30 seconds. Is it ready to submit? No. Is it 70% of the way there and a great starting point? Absolutely. That's the point. You're not eliminating the writer, you're eliminating the blank page.

Donor and stakeholder communications 6:40

The second workflow area is donor and stakeholder communications, and this is where volume becomes the challenge. If you're a development director you might be managing hundreds of relationships: major donors, mid-level donors, corporate partners, board members. Personalization matters, but you can't handcraft every touchpoint.

You can generate a personalized thank-you letter for each giving tier, draft a mid-year impact update that feels warm and specific, or write a lapsed-donor re-engagement sequence. One of my favorites is taking a dry program outcome report and turning it into a compelling donor story in about two minutes.

In the demo I paste in raw program outcome data, the kind of thing that lives in your annual report, and ask AI to reframe it for a donor audience. The key instruction was "lead with a human moment, not a statistic." That is the framing shift that makes donor communications work. We didn't change any of the data. We just changed the entry point.

That becomes your mid-year update, your year-end letter, your board impact report opener. One input, multiple uses. And notice it's not generic. It's specific to Durham, to Title I schools, to that cost-per-student figure. That specificity is what makes donors feel like their money is real.

Board reports and program summaries 10:40

This is one of the most underrated AI use cases I've seen. Executive directors spend enormous amounts of time synthesizing information for board consumption: taking a 20-page program report and turning it into a one-page brief, pulling action items out of two hours of meeting notes, translating raw metrics into a narrative that tells the board what's working and what needs attention.

AI is exceptionally good at summarization and structure. Give it a long document, ask it to extract the key points for a specific audience, and it will do that quickly and well.

In the demo I paste in a rough set of meeting notes, the kind where the bullets make sense if you were there and look like noise if you were not, and ask AI to turn them into a structured action item summary. We get a clean action table with owners, due dates, and decisions captured, in about 20 seconds. Imagine running that after every staff meeting for a year. That is a real operational change.

Content creation 13:20

Content creation is workflow number four, and this is where a lot of nonprofits are already experimenting, but usually in a scattered way. They'll use AI for one-off social posts and never connect it to a broader content strategy.

A more systematic approach starts with one content asset you already have - a grant report, a program description, an impact story - and batches out a full month of content from it: LinkedIn posts, Facebook updates, email newsletter content, Instagram captions. One source, many outputs.

In the demo I paste in a single two-paragraph program description and ask Claude to generate a batch of social content across four channels. What you're looking at is probably two to three hours of content work compressed into less than two minutes.

Does every one of those posts go out exactly as written? Probably not. You'd tweak the voice, add photos, adjust for a current event, change the language a little. But the structure is there. You're editing, not creating from scratch.

Internal operations 15:50

The fifth workflow is internal operations, and this is where smaller nonprofits often find the most immediate relief. It's not grants, not donor communications, just the administrative load that nobody talks about but everybody feels. Meeting follow-ups, staff update emails, onboarding checklists, SOPs. These are all things AI can draft quickly from your raw inputs, and for organizations where one person is wearing five hats, that matters a lot.

In the demo I paste in a rough set of onboarding notes, the kind of thing a longtime employee might keep in their head or in a messy Word doc, and ask AI to turn it into a structured onboarding checklist. The checklist is immediately usable. You would want to add org-specific items, but the structure is there.

Think about all the institutional knowledge sitting in people's heads at your organization right now that has never been documented. AI makes it much easier to get that out of people's heads and into a format the whole team can use and build on.

Building a prompt library 18:00

Everything I've shown you depends on prompts. If you're the only person at your organization who knows those prompts, you've created a bottleneck. A prompt library solves that.

It's just a shared document - a Google Doc, a Notion page, a Word file, whatever your team already uses - organized by category and task, with prompt templates anyone on staff can copy and use.

I'd recommend starting with five categories: grants, donor communications, programs, operations, and social media. Take those five categories, have 10 to 20 prompts for each, and you've built something that multiplies the value of AI across your entire organization.

Mistakes to avoid 19:40

Keep sensitive client data out of public AI tools. No donor financial data or patient information should go into ChatGPT's free tier or Claude's free tier. These are not HIPAA-compliant environments. If you work in healthcare-adjacent programs or handle sensitive client data, talk to your IT provider before using AI for these workflows.

Remember that AI can fabricate. It is called hallucination. It will invent statistics, make up citation sources, and occasionally produce names and information that do not exist. Always verify any specific fact, number, or citation before it goes into a grant or donor communication.

AI drafts are starting points. The most common mistake we see is treating the first output as the final output. It isn't. Your judgment, your relationships, your knowledge of the funder or donor - that's what makes it work. AI can help you get started, but you have to close the gap.

Five takeaways. One, you don't need to be a tech expert; you need the ability to write good prompts and a willingness to iterate. Two, data hygiene matters - know what you're putting in and keep sensitive information out of public tools. Three, start small: pick the one workflow that costs you the most time right now and master it before you expand. Four, a shared prompt library multiplies the value across your whole team. Five, ask questions and practice.