Grant writing is one of the highest-leverage uses of AI for a small UK charity. Done right, AI saves you 5–8 hours per application, surfaces funders you’d miss, and helps you spot the weaknesses in your draft before a panel does. Done badly, it produces generic AI-slop that funders are now trained to filter out. This guide covers the practical ‘done right’ version.

The 4-stage workflow that actually works

Funder discovery — find 10 funders you didn’t know about
Funder analysis — read between the lines of their guidelines
Draft acceleration — get from blank page to v1 in 90 minutes
Critique — make AI play the panel before you submit

Stage 1: Funder discovery

The problem: most charities apply to the same five funders every year. There are 8,000+ trusts and foundations in the UK plus hundreds of council, corporate and lottery routes. You’re missing the ones that fit best.

The AI play: export your last 5 successful applications as a single PDF. Drop it into Claude or ChatGPT with this prompt:

“Based on the funders this charity has successfully won grants from, what characteristics do these funders share — by funding focus, geographic priority, application style, and grant size? Now suggest 10 similar UK funders I should research, with one sentence each on why they’d fit.”

You’ll get a list to cross-check against Funds Online, 360Giving and Charity Excellence. Most will turn out to be irrelevant. A few will be gold.

Stage 2: Funder analysis

The problem: funders publish guidelines, but what they actually fund tells a different story. Reading 20 of their recent grant announcements is the real research, and it takes hours.

The AI play: copy the funder’s last 15–20 grants page (or their 360Giving data) into a single document. Then:

“Here are 20 grants this funder has recently awarded. What patterns can you identify in (a) the size of award, (b) the type of work funded, (c) the geographic distribution, (d) the size of recipient charity, and (e) any surprising themes that wouldn’t be obvious from their published guidelines?”

The patterns you find are signals. A funder publicly “funding all forms of education” but in practice funding only literacy programmes for under-11s in London — that’s the difference between a strong application and a wasted week.

Stage 3: Draft acceleration

The problem: blank-page paralysis. Most charity grant writers spend 60% of writing time on the first 30% of the draft.

The AI play: give the AI three things — (a) the funder’s priorities, (b) your project brief, (c) two of your previously successful applications. Ask:

“Draft a first version of [section] for this application. Match the tone and structure of my previous successful applications. Use my actual project detail. Where you don’t have info, mark [TO FILL] in square brackets.”

You get a v1 that sounds like you, not like AI. You spend the saved hours on the parts that matter — case studies, impact data, the specific numbers and stories the AI can’t invent.

Stage 4: Critique

The most underused step. Once you have a near-final draft, AI becomes a brutal panel reviewer.

“Read this application as if you are a grant panel reviewer at [funder name] with 50 applications to read this week and 4 to fund. Score it 1–10 on clarity, evidence of need, evidence of capability, value for money, and panel appeal. Tell me the top 3 weaknesses that would cause you to skip past it.”

Honest critique you’d normally pay a consultant £400 for. Apply 70% of what it says. Some criticisms will be unfair — your judgement still wins.

What NOT to do

Don’t paste the whole application back to AI for “final polish.” It will smooth out everything distinctive about your voice — exactly what panels notice.
Don’t use AI to invent impact data. Hallucinations get charities blacklisted. Numbers must come from your actual records.
Don’t turn off training without checking. ChatGPT Plus by default doesn’t train on your conversations IF you signed up post-2023; older accounts may need to opt out in Settings → Data Controls.
Don’t paste beneficiary data without anonymising. Replace names and identifiers before pasting personal info.

Recommended tool stack for a UK charity

Claude Pro (£15/month) — for long document work, better British English tone
Funds Online (DSC, £~120/year) — for funder discovery
360Giving data — free, for analysing what funders actually award
Microsoft Copilot — if you’re already on M365, integrates into Word

Total: under £400/year. Easily worth one extra £5k grant secured.

UK GDPR considerations

For data that touches beneficiaries:

Anonymise before pasting — replace names with [Beneficiary 1] etc.
Use EU/UK data residency where available (Resend EU, Claude EU)
Don’t paste medical, criminal or safeguarding details into any AI tool
Have a one-page AI use policy approved by trustees

We help charities through the policy step — including for Camden organisations as part of the subsidised charity half-day session.

One thing to try this week

Pick your next live application. Run just Stage 4 (critique) on your existing draft. 10 minutes. You’ll see 2–3 weaknesses you missed. Fix them before you submit. That single step alone routinely lifts win rates 10–20% for charities we work with.

Help with this for UK charities

We deliver charity-rate AI training for fundraising teams — full detail on the charity digital support page. Free for Camden charities via the Camden pilot.

AI for Grant Writing — A Practical Guide for UK Charities (2026)