How to Use AI in Fundraising: A Consultant Donor Appeal Workflow

Build a verified fact sheet for the client, give the AI only that sheet and a clear brief, have it draft the appeal and impact report, then audit every claim against the sheet before the client signs it off. Keep donor names and giving histories out of the tool, and describe segments in words instead.

The order is the safeguard. Models write fluent fundraising prose very easily, and the fluency is the danger: a draft that says "your gift last year helped 412 children" reads as well when 412 is invented as when it is true. The workflow below makes the facts arrive first, from the client, and treats the draft as something to audit rather than something to admire.

Follow me on Instagram@sagnikteaches

Why a consultant's workflow differs from a charity's

A charity's own staff know their facts, so their risk is time. A consultant works across several clients with different programmes, and the risk is confusion and reputation. You may draft for a food bank in the morning and a youth charity in the afternoon, and the figures must never cross. You are also accountable for what is published under your client's name, even though the client owns the final approval.

Connect on LinkedInSagnik Bhattacharya

That shifts the workflow in three ways. Each client gets its own sealed workspace (a Project in ChatGPT or Claude, one per client). The facts come from the client in writing, with a source and a date. And the final step is a documented approval, which protects both of you. Other tutorials here cover the broader tool choice in AI tools for fundraising consultants and the wider service menu in five jobs a consultant can offer clients; this one is about the appeal itself.

Subscribe on YouTube@sagnikteaches

Step 1: The client fact sheet

The fact sheet is a one-to-two-page document that holds everything the appeal may say. Ask the client to complete it, and refuse to write until it is signed off. For each fact, record the statement, the source, the date and whether it may be used publicly.

An illustrative extract for a charity that funds after-school tutoring, with invented figures:

Two columns do the real work. The date stops stale numbers sneaking into an appeal, and the public-use column stops an internal figure, like the waiting list, ending up in a letter. Also add a short "never say" list: claims the client cannot support, such as "every child improves", or sensitive details about the beneficiaries.

The survey line shows why sample size is part of the fact. "87 of 104" is a claim you can defend; "most parents say" is not, and "84% of parents" quietly drops the base. Put the base in the sheet and require it in the text.

Step 2: Keep donor data out of the tool

An appeal usually needs to speak to different donors differently. You do not need names for that. Write for segments, and let the mail platform insert the personal details.

• Never paste: donor names, addresses, email, phone, gift amounts tied to a person, notes on health or bereavement, and legacy intentions.
• Describe instead: "Donors who have given twice in three years, last gift under $50", "lapsed donors, last gift 18 months ago", "monthly givers on a regular gift".
• Merge in the mail platform: first name, last gift amount and last gift date, with [first name] style fields in your draft.
• Use a business plan, and read what the plan does with your content. ChatGPT Business and Claude Team do not train on business content by default; consumer plans rely on a setting. OpenAI and Anthropic both run nonprofit programmes with reduced seat prices for eligible organisations, so a client may already have an account you can work in, with their permission.

Check your contract with the client, too. Their donor database belongs to them, and a data-processing clause may limit which tools you may use on it. Describing a segment in words keeps personal data out of the tool altogether, which is why the segment approach is the default.

Step 3: The brief and the first draft

With the fact sheet loaded into the client's Project, brief the draft with the same discipline as the sheet: audience, ask, one story, one number, one call to action. Illustrative prompt for a year-end appeal to donors who gave twice in three years:

Write a 330-word appeal letter for donors who have given twice in the last three years (last gift under $50). Use ONLY the fact sheet in this Project. Rules: one number from the sheet in the opening paragraph; the parent quote exactly as written; the ask is $45 to fund one pupil for one term; do not mention the waiting list; no claims that a gift "will" change a life, only what $45 funds. Use [first name] for the greeting. After the letter, list every factual claim in it with the fact-sheet line it came from, and write "UNSOURCED" beside anything you cannot match.

The last instruction is the most valuable one in the prompt. It makes the model do the first pass of the audit for you, and its "UNSOURCED" markers show you where it improvised. A sample reply, illustrative, produces a clean letter and a claims list that matches nine claims to fact lines and marks one UNSOURCED: "children who once struggled now love school". That sentence is plausible, emotional and unsupported, which is exactly the kind of line a human drafting at speed might write too. It is deleted.

Step 4: The claims audit

You still audit, because the model's own source mapping can be wrong. The audit takes ten minutes and works like a proofreader's table. Lay the final text beside the fact sheet and tick each claim.

This is the shape of a real audit: not every flagged item is an invention. "Over 400 volunteers" is a number-mix error, where the model confused pupils with volunteers. These slips are why the claims list exists, and why a human reads it. Also check tenses and time frames: "last year" in a December appeal and "this year" in a January one are different claims.

Step 5: Impact reports from the same fact sheet

An impact report for trustees or a major funder is the same job in a longer form, and the same method applies. Add to the fact sheet the outcomes data, with definitions: what counts as "attended", over what period, and how the survey was run. Ask the model for a structure first, and approve it, then draft section by section.

From the fact sheet, draft the "What we achieved" section of an annual impact report for trustees. Use plain language, 250 words, one table of the figures with their sources and dates. State the survey's sample size next to every survey figure. Do not use words such as "transformed" or "life-changing". At the end, list anything a funder might ask that the fact sheet does not answer.

That final request produces the best output: a list of gaps, such as "no figure for pupils who left the programme", which you then take back to the client. A consultant who arrives with that list at the review meeting looks careful, and the report is stronger for the missing data being found before a funder finds it. For grant applications, the step-by-step grant writing tutorial uses the same fact-sheet approach, and what funders think of AI use is worth reading before you promise anything.

Step 6: The client approval step

Nothing leaves the building without the client's written approval. A simple one-page form does the job, with a line for each of these:

1. I have read the final text in full.
2. Every figure and quote matches our records, or is marked for correction.
3. Any beneficiary story or quote has current written consent.
4. I understand AI tools were used in drafting, and that the text was audited against our fact sheet.
5. Name, role, date, and the version number of the text approved.

Version control matters. If a board member changes a number the day before sending, the approval no longer covers the new text. Put the version in the file name and ask for approval of that file. A signed form also gives the client's trustees comfort when they ask how AI was used, and it keeps the question of responsibility clear.

A full run: a year-end appeal in a week

Here is the whole workflow in time, for a consultant with an illustrative three-client retainer and a year-end appeal for the tutoring charity.

That is about 5.5 hours of consultant time for three tested segment letters and follow-ups, against perhaps 10 to 12 when each is written from scratch. These are illustrative; time yourself on the first two clients before pricing anything on them. The fact sheet takes the client more time than any other step, and that is by design: a client who cannot fill it in is a client whose appeal would have been built on guesses.

Before and after: an appeal opening, audited

This is the same opening twice, first as a fast, unaudited draft and then after the workflow. Both are illustrative.

Unaudited: "Every day, children across our community struggle to keep up at school. Thanks to supporters like you, over 400 volunteers gave their time this year and transformed hundreds of young lives."

After the audit: "Last school year, 412 pupils came to our after-school sessions at least once. This term, $45 pays for one pupil to join for a full term, with a volunteer tutor beside them."

The first version has four problems you can name: an unsupported claim ("every day, children across our community struggle"), a wrong number (over 400 volunteers against 38), an outcome nobody measured ("transformed") and no ask. The second is shorter, uses two facts from the sheet, and makes a specific request. It is also less emotional, which is where the story comes in: the parent's quote, exactly as consented, belongs in the second paragraph, and does more with eight true words than the first version did with a hundred unsupported ones.

Thank-you and follow-up messages from the same sheet

Donors who hear nothing after a gift lapse more often, so the thank-you sequence deserves the same care. It is also a safe place to start with a client who is nervous about AI, because the content is gratitude and a fact or two.

Draft three short emails for a donor who has just given $45 online: (1) an immediate thank-you, (2) a three-week update, (3) a year-end note. Use only the fact sheet. Each under 120 words. The first states what $45 funds, the second gives one figure from this term, the third invites a second gift without pressure. Use [first name] and [gift amount] merge fields. Do not describe the donor's motives or feelings.

A sample result, illustrative: the immediate email thanks [first name] for the [gift amount] gift, says that $45 pays for one pupil's tutoring for a term and promises an update in three weeks. What you fix is usually the promise: if the client cannot send an update in three weeks, the line must change, because a donor who is told to expect something and does not receive it loses trust faster than one who was never told. Ask the client what they can honestly commit to, and write that.

A second client: the same method on a food bank

To check the workflow is not tied to one charity, run it on a very different one. A small food bank wants a winter appeal. The fact sheet shows 2,140 parcels packed since October, a stock room that holds eight weeks of tinned goods, and a cost of $14 per parcel including transport. In the first draft, the model writes "a parcel feeds a family for a week". The fact sheet does not say that; the food bank's own guidance says a parcel covers three days of meals for a household of two. The claims list marks it UNSOURCED, and the corrected line reads "one parcel, $14, covers three days of meals for a household of two".

This is a realistic mistake, and a revealing one: the model's version sounded better and overstated what a parcel covers more than twofold. The audit turns that into a catch rather than a complaint from a supporter. It also shows why per-client Projects matter. If the food bank's "three days" ever found its way into the tutoring charity's letter, you would have a different, subtler error, and only separate workspaces prevent it.

Tone and style: a quick rubric

Accuracy is the first test. The second is whether the appeal sounds like the client and treats donors and beneficiaries with respect. Score each draft from 1 to 5 on the following, and reject anything under 4 on any line:

• Specific: does it ask for a stated amount for a stated purpose?
• Honest: would the finance director and the programme lead both sign off every sentence?
• Dignified: does it avoid pity, shock images in words and any suggestion that beneficiaries are helpless?
• Donor-centred: does it say what the donor's gift does, not just what the charity does?
• In the client's voice: would a trustee believe the director wrote it?

For the last line, feed the client's best past letters into the Project as voice examples, with their permission. Ask the model to describe the voice in five points first, and check the description matches your own reading before it writes anything.

Consent for stories and photographs

Stories raise money, and they are the point where AI use and ethics meet most sharply. AI can tidy a story the client has collected: fix grammar, shorten it, match its length to the format. It must not create a story, merge two into one or add detail the person did not give. Written consent for the story should state the wording or its purpose, where it will be used, and how the person can withdraw. For children and vulnerable people, a guardian's consent and the client's own safeguarding rules apply, and a client's safeguarding lead should see the final version.

The same applies to AI-generated images used in appeals. An image of a "child at a desk" produced by a model is not a photograph of the charity's work, and presenting it as one misleads donors. If an image is illustrative, label it as such or use the client's own photographs with consent. Some sectors have codes of practice on honest fundraising; ask the client which apply to them and add them to the audit list.

Where AI-assisted appeals go wrong

The invented beneficiary. A draft includes a composite child, complete with a first name and an age, who does not exist. Never allow named or described individuals unless they are on the fact sheet with consent.

The borrowed claim. The model brings in a sector statistic, such as a general claim about children and tutoring, with no source. Add "no outside statistics" to every prompt and delete any that survive.

The stale figure. The sheet says 412 and the client has since added a late cohort. Date every fact and refresh the sheet before each campaign.

The over-promise. "Your gift will change a life" claims an outcome nobody can guarantee. Prefer "$45 funds a term of tutoring for one pupil", which is what the money does.

The same letter to everyone. The segments blur and the appeal reads as generic. Write one brief per segment, and read the three letters side by side to confirm they differ in more than the greeting.

Keeping the workflow safe as it scales

With five or ten clients, the risk moves from a single draft to the system. Keep one Project per client, with a naming rule, and never reuse a Project across clients. Store the fact sheets, claims audits and signed approvals in the client's own folder. Review once a quarter: sample two appeals per client, re-run the audit and see whether any claim crept in after sign-off. For donor-facing media beyond text, such as updates in video, the tutorial on donor update videos and disclosure rules covers what to tell donors.

For templates you can hand a client directly, the donation request letter template with AI gives a simpler starting point. And if you would like to plan this for your consultancy, my AI implementation consultation covers it.

Consultant questions about AI appeals

Can I put a donor list into ChatGPT to personalise appeals?

Not names and giving histories. Describe the segment in words (for example, donors who gave twice in three years) and write the appeal for the segment. Merge the personal fields, such as first name and last gift, in your mail platform, where the data already lives under the client's terms.

Should I tell the client I use AI?

Yes, in the proposal and contract. Say which stages use it, that every claim is checked against a client-supplied source, and that the client approves the final text. Some funders and boards ask, so a clear line protects the client as well as you.

How do I stop the AI inventing impact numbers?

Give it a fact sheet and forbid anything outside it. Ask it to mark any claim it cannot source from the sheet, then run a claims audit afterwards: every number and quote in the final text must match a line on the sheet or be deleted.

Can I use real beneficiary stories?

Only with written consent from the person, or a guardian where relevant, covering the wording you will use. AI can tidy a story the client has supplied, but it must never write one. Composite or invented stories presented as real undermine trust and may break advertising rules.

Does this work for a one-off campaign and an ongoing retainer?

Yes. Build the fact sheet and prompts once per client, refresh the sheet each campaign and keep the audit step every time. On a retainer the fact sheet becomes a living document, which is itself a service the client values.

Further reads

• Can AI Write a Sports Club Grant Application? What Funders Expect — What funders expect from an AI-assisted grant application.
• AI Grant Writing Tools Compared for Small Non-Profits — Grant-writing tools compared for small charities.
• OpenAI for Nonprofits: Who Qualifies and What to Check First — Who qualifies for OpenAI nonprofit pricing.
• AI Tool Discounts for Charities and Non-Profits: What to Claim — Nonprofit discounts on AI tools worth claiming.
• AI Consultant for Nonprofits: What to Hire For and What to Pay — What a nonprofit should expect to pay for AI help.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: OpenAI ChatGPT Business pricing page (including its nonprofit FAQ), Claude pricing page, Claude for Nonprofits page. Check current terms and your client's data agreements before use.