Charity Appeals and Social Posts With AI That Keep Your Voice

Give the AI your voice before you ask for any copy: three to five past appeals that raised well, a one-page voice card (words you use, words you never use, how you describe the people you help) and one real, consented story. Then ask for drafts, and edit every one for facts, dignity and rhythm before it's published.

Without that groundwork, AI fundraising copy drifts to the same phrases every other charity's AI draft uses: "make a real difference", "together we can", "your vital support", "now more than ever". Supporters have read them hundreds of times and skim straight past. A charity's voice isn't a set of adjectives. It's the specifics: who you helped, what exactly happened, what the gift buys, and how you talk about people without talking down to them.

Follow me on Instagram@sagnikteaches

Pulling your voice out of appeals that already worked

Start with evidence, not a brainstorm. Find three to five pieces that raised well or drew replies: a winter appeal letter, an email that got unusually high clicks, two social posts people shared. Paste them into a chat assistant with this prompt:

Connect on LinkedInSagnik Bhattacharya

These are five fundraising pieces from our charity that performed well.
Describe the voice they share, using evidence from the text:
1. Sentence length and rhythm (quote two typical sentences)
2. How we describe the people we help (exact words we use)
3. How we make the ask (quote each ask)
4. Words and phrases we repeat
5. What we never do (e.g. exclamation marks, guilt, jargon)
6. Anything inconsistent between the pieces
Don't praise the writing. Just describe it.

For a food bank, the reply might read (illustrative):

Subscribe on YouTube@codingliquids

1. Short sentences, often under 12 words. Paragraphs of 1-2 sentences.
   "[First name] is seven. She asked her mum if they'd have dinner tomorrow."
2. "Families", "neighbours", "people using the food bank". Never "the needy"
   or "vulnerable". Ages given, surnames never.
3. Asks are specific and small: "$25 fills a bag for a family for three days."
4. "Neighbours", "this winter", "on the shelves".
5. No exclamation marks. No "please help". No statistics about poverty in general.
6. The March email uses "clients"; the others say "families".

Point six is useful on its own: the team hadn't noticed they'd drifted between two words. Decide which one you mean and put it on the card.

A filled-in voice card for a food bank

Turn the analysis into one page the AI (and new volunteers) can follow. Here's a complete example for an illustrative community food bank, written in the form I'd paste into a shared project:

VOICE CARD: [Food bank name]

Who we are in one line: Neighbours making sure nobody in town goes without food.

We sound like: a volunteer telling a friend what happened at the food bank today.
Plain, warm, specific. Short sentences. No exclamation marks.

People we help: "families", "neighbours", "people who use the food bank".
First names only, changed unless we have written consent. Ages are fine.
Never: "the needy", "the vulnerable", "the less fortunate", "beneficiaries".

What gifts do (from our records, update each quarter):
$25 = a three-day food bag for a family of four
$60 = a week of breakfast club for 20 children
$150 = the electricity to run our fridges for a month

How we ask: once, clearly, near the end. One amount, one thing it buys.
Never: guilt, "please help", "only you can", invented deadlines.

Words we like: shelves, bag, winter, neighbours, thank you.
Words we cut: vital, crucial, heartbreaking, make a difference, now more than ever.

Every post gets checked for: real facts, consent, dignity, one clear ask.

The "what gifts do" figures are the most valuable lines on the card and the ones AI must never supply itself. They come from your finance records, and someone should own updating them. The general method behind a card like this is in building a brand voice guide that AI can follow.

Store the card, two or three of your best past pieces and your current gift figures in a ChatGPT or Claude Project (both offer shareable Projects on business plans) or a Gemini Gem (becoming a skill from November 2026), so every person drafting starts from the same material instead of their own prompt.

Briefing one appeal: the story, the ask, the proof

Most flat AI appeals come from flat briefs. "Write a winter appeal for our food bank" gives the model nothing specific, so it fills the gap with stock phrases. A brief that works has three parts: one real story, one ask, and one piece of proof.

Using our voice card, draft a winter appeal email (250-300 words).

Story (consent: written, first name changed, photo not allowed):
[first name], 41, lost his warehouse job in September. Two children, 9 and 12.
First visit in October. Said: "I walked past three times before I came in."
Now volunteers on Thursdays sorting donations.

Ask: $25, which fills a three-day bag for a family of four.
Proof: 312 families came to us last winter; the shelves ran empty twice in January.
Target: $18,000 by 31 December.

Don't add any facts, figures, quotes or details that aren't in this brief.
Give me two subject lines and flag any sentence you're unsure about.

The last two lines of the brief do more for accuracy than anything else. Asking the model to flag uncertain sentences surfaces the places it was tempted to invent.

The subject lines deserve the same scrutiny as the body. From this brief, the model might offer (illustrative):

1. "I walked past three times before I came in"
2. "Don't let a family go hungry this Christmas"

The first is the father's own sentence, it's specific, and nobody else's appeal can use it; keep it. The second breaks the card twice. "Don't let" is the guilt the card rules out, and "Christmas" appears nowhere in the brief: the model assumed it from the 31 December target, and a food bank serving families of several faiths may not want every winter appeal framed that way. A better second option, built from the proof line: "The shelves ran empty twice last January".

Consent needs its own record, because the brief's one line ("written, first name changed, photo not allowed") has to hold for every piece of the campaign. A filled-in entry for the story above:

STORY: [first name] (changed)       Consent taken: [date], by [staff initials]
Agreed channels: appeal email, Facebook, Instagram, newsletter
Not agreed: photo of him or his children, LinkedIn, press
Details allowed: age, job loss in September, children's ages,
  "walked past three times", volunteers on Thursdays
Details not allowed: which warehouse, the children's school
Use again after 31 December? Ask him first

If he changes his mind on day 9, the record tells you what to take down: delete the Facebook and Instagram posts, drop him from the day-14 email, and note the withdrawal date. The launch email has already been sent and can't be recalled, which is worth telling him honestly when you confirm the rest is gone.

Before and after: the first draft and the version that went out

Even with a good brief, first drafts need an editor. Here's an opening paragraph as it typically comes back (illustrative):

This winter, families across our community are facing impossible choices. [First name], a devoted father of two, never imagined he would need help. But when he lost his job, everything changed. Your generous gift today can make a real difference to families like [first name]'s, providing vital support when they need it most.

And the edited version:

[First name] walked past our door three times before he came in. He'd lost his warehouse job in September. He has two kids, nine and twelve. "I didn't know what I'd say," he told us. What he got was a bag of food, a cup of tea and nobody asking questions. These days he's back on Thursdays, sorting the donations you send.

What changed: "impossible choices" and "never imagined" were the model's inventions, so they went. "Devoted father" is a judgement the charity can't make. "Make a real difference" and "vital support" came off the cut list. The edit leads with the detail from the brief that nobody else could write (three times past the door) and saves the ask for later. That's the work: keep the model's structure, replace its phrases with your facts. For more side-by-side edits like this, see tone of voice examples before and after editing.

One appeal, a fortnight of social posts

Once the appeal email is approved, it becomes the source for everything else. Ask for a posting plan built only from the approved text, so nothing new creeps in:

A day-3 Instagram draft might come back as: "Empty shelves mean empty plates 💔 Just $25 can fill a bag for a family this winter. Every donation counts! Link in bio 🙏". It breaks the card three ways: exclamation mark, a slogan instead of a detail, and emojis the charity never uses. The fixed version: "This is one bag. Pasta, tins, cereal, tea, nappies if they're needed. $25 fills it, and it feeds a family of four for three days. Link in bio." Before anything is scheduled, run it through a short check like the pre-publish checklist for AI-drafted posts.

The day-5 LinkedIn post goes wrong differently, because the model shifts into corporate language for a business audience. An illustrative draft: "Partnering with [food bank] is a powerful way to demonstrate your commitment to social responsibility. Twelve local businesses have already joined our breakfast club sponsorship scheme. Join them and make an impact." Nothing in the brief mentions twelve businesses or a scheme. The fixed version goes back to the card: "Our breakfast club feeds 20 children before school, five days a week. $60 keeps it running for a week. If your business would like to sponsor a week this winter, message us and we'll send you a photo of the tables that morning." A concrete, bounded offer gives an employer something to say yes to; a pitch about their values doesn't.

The day-8 progress post is where numbers go wrong. Say $7,450 has come in against the $18,000 target. Asked for a progress update, the model writes "we're already more than halfway there!", which is false: $7,450 is about 41%, and the exclamation mark breaks the card. Do the sum yourself and give the model the result: $7,450 divided by $25 is 298 bags. "You've filled 298 three-day food bags so far. We're at $7,450 of $18,000, with [X] days to go." Every figure in that sentence can be checked against the donations report before it's posted.

Lines AI copy must not cross in charity fundraising

A business can get away with slightly generic marketing. A charity trades on trust, and the ways AI can damage it are specific:

• Invented or composite people presented as real. If the model adds a quote, an age or a detail that wasn't in the brief, it's fiction. If you ever use a composite story to protect someone, say so in the copy.
• AI-generated images of people you help. In 2023 Amnesty International withdrew AI-generated images from a campaign about police violence at protests after photojournalists criticised them, even though the aim was to protect real protesters' identities. For a small charity, a photo of the shelves or a consented, anonymised photo is safer than a lifelike generated face. If you sell to or campaign among people in the EU, the AI Act's labelling duty for AI-generated and deepfake content has applied since 2 August 2026.
• Figures the model supplies. "One in five families in our area…" from an AI draft is a claim you now have to defend. Every number comes from your records or a named source, or it comes out.
• False urgency. "Only 48 hours left to save our service" when there's no deadline is a lie, whoever typed it.
• Stories without consent. Consent should cover the channel. Someone who agreed to a newsletter mention may not have agreed to Instagram. Children's stories need a parent's or guardian's consent and extra care with identifying details.

If supporters ask whether you use AI, be straightforward about it: it drafts, people write and check. Whether to tell people you use AI in marketing goes through the options.

Phrases to strip from any AI fundraising draft

Keep this list on the voice card and search every draft for it before editing anything else. Each has a better replacement, which is always something specific:

• "Make a real difference" → say what the difference is: "keeps the breakfast club open for a week".
• "Vital work" or "crucial support" → name the work: "Thursday sorting", "the Tuesday drop-in".
• "Vulnerable families" → describe the situation: "families where a parent lost work this autumn".
• "Now more than ever" → give the fact behind it, or cut it: "we ran out twice last January".
• "Together, we can…" → cut it; the ask already implies it.
• "Heartbreaking", "devastating", "impossible choices" → let the story carry the feeling without the label.
• "Your generosity" in every paragraph → once, at the end, as thanks.

How the same method changes for other kinds of charity

The food bank is one example. The card, brief and edit stay the same elsewhere, but what goes on the "never" list changes with the cause:

• A hospice-at-home charity. Stories involve people at the end of life and their families, so consent is often given by relatives and should be confirmed again before each new use. AI drafts tend to reach for "brave battle" and "lost their fight"; many families dislike both. The voice card might say: describe what the nurse did, not how the patient "fought". A draft line such as "After a brave battle, [first name] passed away peacefully at home" becomes "Our nurse sat with [first name]'s family through the last night, so they could all be at home together", with the family's agreement to that exact wording.
• An animal rescue. The animal is the story, which makes AI copy prone to cute, human-voiced clichés ("Bella just wants a forever home!"). A rescue with a drier, practical voice should put "no speaking as the animal" on the card, and check adopters' consent before using their photos or street names.
• A youth sports charity. Children's first names, school names and training times together can identify a child. The card should ban all three in the same post, whatever the AI suggests, and every photo needs a parent's or guardian's written consent for that channel.

Who writes, who checks, who presses publish

In a small charity three or four people might post: the fundraiser, the chief executive, a volunteer who looks after Instagram. Voice drifts fastest when each uses their own prompt in their own account. Two rules keep it steady. First, everyone drafts from the same shared project with the voice card in it, never from a blank chat. Second, anything with a beneficiary story, a figure or an ask gets a second pair of eyes before publishing; a thank-you or event photo can go straight out. Write those two rules into your AI policy so new volunteers learn them on day one rather than after a mistake.

When the campaign closes, write one line per piece in the same document as the posting plan: what it raised through its tracked link, or the replies and shares it drew. Illustratively, "Day 3 Instagram, shelves photo: 41 gifts; day 10 volunteer post: 2 gifts, 30 comments". The shelves post raised money and the volunteer post built goodwill, and both are useful to know. The pieces that performed become the examples in next quarter's project, which is how the voice card keeps learning from your supporters rather than from the model.

Review the card every quarter. Update the gift figures from the accounts, add any phrase the team keeps cutting, and swap one of the example pieces for your latest best performer. A card that's a year old starts producing last year's voice.

With the voice card in place, a food bank's winter appeal might go from half a day's writing to about 90 minutes: 20 minutes to write the brief, 10 for drafts, and an hour of real editing. The hour of editing is the part you can't skip, and it's where your voice actually comes from.

Further reads

• Donor Thank-You Letters With AI That Still Feel Personal — The same voice work applied to thank-you letters.
• How to Use AI to Write Grant Applications, Step by Step — A different reader: funders rather than supporters.
• How to Write Instagram Captions With AI That Sound Like You — More on captions that don't read as AI.
• How to Write an AI Policy for a Small Charity — Write your rules on AI images and stories into policy.
• Turning Charity Data Into Impact Reports With AI — Where the proof points in your appeals should come from.
• Should a Small Business Use AI to Write Its Social Media Posts? — When AI-drafted posts help and when they hurt.
• Using AI to Spot Donors Who Are About to Lapse — A step-by-step method for flagging donors who are drifting, using a pseudonymised export, an overdue-ratio score and a person checking the list.
• How Small Charities Can Use AI With Almost No Budget — A seven-stage plan for a charity with no AI budget: get verified, pick a free workspace, choose three jobs, set a data rule, and trial it for a month.
• Can You Use AI for Grant Writing Without Funders Rejecting It? — What funders have said about AI-assisted applications, why generic drafts score badly, and a worked $15,000 application that stays in your own voice.
• How to Write Case Studies From Customer Interviews Using AI — Interview questions, transcript-to-brief and draft prompts, and an approval routine for case studies where every result comes from the customer, not the model.
• How to Write LinkedIn Posts With AI Without Sounding Like AI — Start from a two-minute voice memo, let AI tighten without rewriting, and remove the tells readers spot in a scroll, with a before-and-after post and hooks.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: EU AI Act Article 50 transparency obligations; news coverage of Amnesty International withdrawing AI-generated campaign images (May 2023); ChatGPT and Claude Projects help pages.