Claude for grant writing: a safe workflow for nonprofits

AI is good at grant writing in one specific way: it turns approved material into a structured first draft that fits a funder’s template. It is bad at knowing facts you haven’t given it. Most of the risk, and most of the time saved, comes from how well you handle that gap.

This page sets out the workflow we use when we build proposal and report drafting for nonprofits on Claude. It assumes you already have a Team plan; if not, our Claude for Nonprofits guide covers pricing, eligibility and set-up. Anthropic’s nonprofit plugin also ships fundraising skills for finding funders and writing proposals and reports. The workflow below sits around those skills, whichever you use.

The lesson that changed how we build this

When we built a proposal-drafting skill for a livelihoods training nonprofit this year, the most useful request came from the client, not from us. They asked for the source content as an editable document so the fundraising team could correct outdated programme details.

That changed the design. The reference pack became something the team owns and updates, rather than text buried inside a tool. The broader point: if your proposals pull from stale material, Claude will produce confident, stale drafts faster. The reference pack matters more than the prompt.

The workflow, step by step

Step 1: Build the reference pack. In a Claude Project, load approved, current material only:

• programme descriptions and theory of change;
• standard organisational text (history, governance, registrations such as 12A/80G/FCRA where relevant);
• budget templates and logframes;
• two or three past proposals that were funded;
• your latest annual report figures, taken from your system of record.

Name one owner, usually the fundraising lead. Keep the master copy as an editable document outside Claude so it can be corrected and reloaded.

Step 2: Write the project instructions. Tell Claude to ask clarifying questions before drafting, to use only facts in the reference pack or the funder brief, and to mark anything it cannot source with a visible flag such as [SOURCE?]. Ask it to follow the funder’s headings and word limits exactly.

Step 3: Add the funder brief. Paste or upload the call for proposals, the template and the scoring criteria. If your organisation tracks opportunities in grant management software, export the funder’s requirements from there so deadlines and reporting conditions stay in one place.

Step 4: Draft section by section. Ask for one section at a time, not the whole proposal. You get better drafts and it is easier to see where Claude needed facts it didn’t have.

Step 5: Review against a checklist. The reviewer (named in your nonprofit AI policy) checks:

• every number against the system of record;
• every claim about outcomes against evidence you hold;
• no invented quotes, case stories or partner names;
• every [SOURCE?] flag resolved or removed;
• the funder’s eligibility rules and word limits met;
• tone and language right for this funder.

Step 6: Feed corrections back. Anything the reviewer fixed that came from outdated material goes back into the reference pack. This is the loop that makes the next proposal faster.

The same workflow for funder reports

Quarterly donor reports and CSR utilisation narratives follow the same pattern with one change: the numbers come in first. Export figures from your MIS or programme database, then ask Claude to write the narrative around them, align it to the funder’s indicators and flag gaps (“the template asks for disaggregation by gender; the export has none”). Claude should never be the source of a figure.

What to keep out of grant drafts

• Beneficiary personal data. Case stories in proposals should be consented and de-identified before they go anywhere near Claude.
• Donor bank or payment details. No reason for them to be in a drafting workflow.
• Unverified outcome claims. Funders are getting better at spotting AI-polished claims that don’t match reported data. It is a trust problem, not a style problem.

Ready to put Claude to work in your organisation?

Grant drafting is often the second workflow we set up, after an internal knowledge base. Book a Claude rollout planning call with EdZola. Bring your next proposal or funder report and we’ll map how Claude fits, what your team needs to review and what it will cost to run.