AI can draft booking replies, volunteer rota messages, newsletters and the first version of a grant application, as long as a named person checks each one before it goes out. It cannot take safeguarding decisions, handle complaints or accident records, sign off safety checks, or decide who gets a room. Those stay with people.
What separates the two lists is the cost of a wrong sentence. A wrong sentence in a newsletter costs a correction; a wrong sentence in an incident record or a safeguarding note can harm someone and cannot be quietly fixed later. The jobs people most want to hand off, such as the awkward complaint or the incident write-up, are often the ones that most need a person.
Four tests that decide whether a job can go to AI
Ask four questions about any job. Could a wrong sentence hurt someone, such as a child, a volunteer or a hirer? Does it commit the centre's money, space or promises beyond a limit your committee has agreed? Does it involve private information about a named person, such as health, behaviour or a safeguarding worry? Does it change who may be in the building or what may happen there, as keys, licences, insurance conditions and fire safety do? If every answer is no, the job can go to AI as a draft with a named approver. One yes keeps the decision with a person, and two yes answers mean the assistant should not see the material at all.
| Job | Tests it fails | Who does what |
|---|---|---|
| Reply to a hire enquiry | None, if the reply quotes only your own documents | AI drafts, the booking clerk approves |
| Confirm a booking and take a deposit | Money | AI may draft the wording, the manager confirms |
| Volunteer rota messages | None | AI drafts, the coordinator sends |
| Build the rota grid | Who is in the building (clearances) | AI fills the grid, the coordinator checks every shift |
| Monthly newsletter | None | AI drafts from notes, an editor approves |
| Grant application | Promises to a funder | AI drafts sections from your facts, a trustee or manager signs |
| Safeguarding concern | Harm, private information | A person, following your policy; no AI |
| Complaint about the centre or a volunteer | Private information, possible harm | A person, following your complaints procedure |
| Accident or incident record | Harm, private information | A person writes it at the time; no AI rewriting |
| Fire and safety checks | Harm, who is in the building | A competent person inspects and signs |
| Two groups want the same room | Space, money | The manager decides against your allocation policy |
A centre with three rooms: where the hours go
Consider a centre with a main hall, a meeting room and a kitchen, 14 regular groups, about 35 booking enquiries a month, 22 volunteers, a monthly newsletter to 600 subscribers and two grant applications a year. The minutes below are assumptions; replace them with your own stopwatch readings.
| Job | How often | Minutes now | Minutes with an AI draft | Saved per month |
|---|---|---|---|---|
| Booking enquiry replies | 35 a month | 6 each | 3 each | 105 |
| Volunteer rota messages | Weekly | 45 | 20 | 100 |
| Newsletter | Monthly | 180 | 75 | 105 |
| Grant first drafts | 2 a year | 480 each | 300 each | 30 (360 a year) |
| Total | 340 |
That is 5 hours 40 minutes a month, about an afternoon. It is worth having for a volunteer-run committee, and it does not replace anyone. The jobs in the person-only half of the table take less time but matter more, and none of that time is meant to shrink. If the hours AI frees go into a safeguarding refresher or a proper chat with a new volunteer, the tool has done something useful.
Booking replies: a draft that quotes only your own documents
Most enquiries ask the same few things: is the room free, what does it cost, what are the rules. The risk is an assistant that answers from habit instead of from your terms. The fix is a prompt that pastes in your rate card, hire terms and a calendar extract, and forbids anything outside them.
You draft replies to hall-hire enquiries for a community centre. Use
only the rate card, hire terms and calendar extract below. If something
is not covered, say a member of the team will confirm it and put the
question on a line beginning "For the manager:". Never promise a
discount, a date missing from the calendar extract, or that alcohol or
amplified music is allowed. Under 120 words, friendly, first name only.
RATE CARD: [paste]
HIRE TERMS: [paste]
CALENDAR EXTRACT: [paste the free slots for the dates asked about]
ENQUIRY: [paste with surname and phone number removed]An enquiry might arrive like this: "Hi, we're local residents and would like the main hall on a Saturday afternoon in mid-November for a child's birthday party, about 30 people. What does it cost?" With the rate card (private bookings at $40 an hour, community groups at $28, kitchen $10 extra, $50 refundable deposit) and the calendar extract showing 2 to 6 p.m. free, the draft should quote the private rate for four hours, $160, and the deposit, confirm the slot from the calendar and ask about food and decorations. It should add, "For the manager: confirm adult-to-child ratio for a children's party."
A first draft can offer "our community rate", because the enquirer wrote "local residents". The rate card reserves that rate for registered groups, so add one sentence to the hire terms: "The community rate applies only to groups on the register." The assistant then applies the rule it can see. Never let it confirm availability on its own; the calendar extract is the only source it may use, and a person checks the live calendar before replying. The tutorial on village-hall booking systems and AI replies shows how to wire the replies into a booking tool. Refunds above the deposit, damage claims, and anything touching alcohol, late music or licences go to the manager, not to a draft.
An enquiry about an 18th birthday with a DJ and a bar is the line the prompt must not cross. The right draft is short: "Thanks for asking, [first name]. Events with music or alcohol need a conversation with the manager first, because of the centre's licence and insurance conditions. I have asked them to call you on [number] by [day]." That tells the hirer what happens next and decides nothing. Test the prompt on five awkward enquiries before it sees a live one: a late finish, a request for a discount, a hirer asking to store equipment, a group wanting its own key, and a cancellation after the deposit. Each should end in a "For the manager" line, not an answer.
The same documents produce hirer reminders. A message three days before the booking can list the start and finish times, who hands over the key, the cleaning checklist and how the deposit is returned, all drawn from your hire terms. Keep key handling and entry codes out of the assistant entirely. The draft says "the manager will send access details separately", and a person sends them.
Volunteer rotas: AI fills the grid, a person owns safeguarding
Rota building is a puzzle with rules, which assistants do reasonably well when the rules are explicit. Use volunteer codes in the data, never names, and tell the assistant to leave a gap instead of guessing.
Draft next week's volunteer rota from the availability table below.
Shifts: Tue lunch club 11-2 (2 people); Wed youth club 5-8 (2 people,
at least 1 marked YC); Thu coffee morning 9-12 (2 people); Sat hall
opening 8-10 (1 person marked KEY).
Rules: nobody works more than two shifts; fill each shift only from
people marked available; if a shift cannot be filled, say so instead
of assigning someone unavailable. Use volunteer codes only.
Availability: [paste table with codes, days available, YC and KEY marks]A result might read: Tue lunch club V03 and V07; Wed youth club V02 (YC) and V09; Thu coffee morning V03 and V11; Sat hall opening V05 (KEY). It should add: "No gaps. V03 has two shifts. I cannot verify the YC marker; it comes from your table." That last line is the point. The YC mark stands for a volunteer cleared to work with young people, and only your coordinator holds the vetting records. Keep the vetting data out of the prompt, keep the mark in your own table, and have the coordinator confirm every youth shift before the rota goes out.
The message that follows is the easy part. Before: "Hi all, rota attached, shout if you can't do your slot, thanks!" After: "Hi [first name], you're down for Tuesday's lunch club, 11 to 2, and Thursday's coffee morning, 9 to 12. If you can't make either, please reply by Sunday evening so we can find cover. Thanks, [coordinator]." The second version names the shifts, gives a deadline and says who to tell. For volunteers who simply choose slots, a no-AI tool may beat both. SignUpGenius's pricing page lists a free plan with unlimited sign-up pages and automatic confirmation and reminder emails, Silver at $8.99 a month, and Gold at $22.49 with waitlist management and ten administrators. Silver's list begins with "1 ad-free sign up", so free pages probably carry ads; check what your volunteers will see. The tutorial on volunteer forms and rota reminders builds the form version.
Newsletters: rough notes in, a send-ready draft out
A newsletter is the safest big job for AI because every fact in it comes from you. Feed it your notes, not a blank page. Suppose the notes say: "Hall closed 23 to 27 Dec. Lunch club moves to Wednesdays from January. Need 3 more volunteers for coffee mornings. Thanks to the knitting group for 40 hats. Reminder: no parking on the forecourt during youth club." Ask for a friendly 250-word newsletter with a short subject line, using only those facts, and a list at the end of anything it was unsure about.
What to check in the draft: dates and days against your calendar, the number of volunteers wanted, and any detail the assistant added. A common slip is turning "from January" into "from 1 January", a date nobody decided. It will also decorate: "40 beautiful hats" is harmless, but "the committee is delighted to announce" puts words in your mouth. Photos of children, names of volunteers and any detail from a member's circumstances go in only with consent. Free email plans may not cover a list of this size: Mailchimp's Free plan allows 250 contacts and 500 sends a month, and MailerLite's allows 250 subscribers and 2,500 emails, so a 600-address list needs a paid tier. Mailchimp lists paid plans from $13 a month and MailerLite from about $11 a month on yearly billing, both for around 500 contacts, so check the price for your count.
Summaries and minutes: AI organises, a person checks the facts
A funder's guidance can run to 40 pages when you need the dates, eligible costs, exclusions and attachments. A document that size sits well within ChatGPT's upload limits, which OpenAI lists as 512 MB and 2 million tokens a file. Upload it and ask for a checklist with page numbers, so every line can be checked against the original.
Read the attached funder guidance. Produce a one-page checklist: closing
date and time, maximum grant, who may apply, eligible and ineligible
costs, required attachments, scoring criteria, and any rule about the
use of AI. Put the page number after each item. If the guidance does
not say, write NOT STATED instead of guessing.The checklist might come back as: "Closing date: noon on 14 February (p. 3). Maximum grant: $10,000 over 12 months (p. 4). Ineligible: building purchase, retrospective costs (p. 6). Attachments: governing document, latest accounts, safeguarding policy (p. 9). Use of AI: NOT STATED." Check the closing date and time against the page cited before you plan around it. Assistants sometimes merge two deadlines, such as an expression of interest and the full application, which is why the page numbers matter.
Meeting notes follow the same pattern. Paste your own rough notes and ask for draft minutes under standing headings: apologies, decisions, actions with an owner and a date. The chair checks and signs. Keep confidential items (staffing, safeguarding, anyone's circumstances) out of the notes you paste, and write those up separately by hand.
Grant drafts: what AI may write and what must be yours
Grant writing splits cleanly. AI can turn your facts into tidy paragraphs, check a draft against a funder's criteria and cut a 1,200-word answer to 800. It cannot supply the facts, and it will invent plausible ones if you let it. The rule I would put in writing: every number in an application must trace to a record you can show, and the assistant is told so.
Before: "Our community centre serves thousands of residents and is a vital hub for the local community." After, with figures from the booking log and the lunch club register: "In the last 12 months the centre hosted 214 bookings from 14 regular groups, including a weekly lunch club for about 30 older residents and a youth club on three evenings a week." Mark each figure with its source in square brackets while you work, such as [booking log, Oct to Sep], and delete the brackets before sending. The second version is more convincing because a funder can check it.
The second job is the criteria check. OpenAI's own page for nonprofits suggests uploading a draft with the funder's evaluation criteria and asking for an assessment. A prompt for it:
Below are a funder's assessment criteria and my draft answer. For each
criterion, give a score from 1 to 5, quote the sentence in my draft that
supports the score, and say what evidence is missing. Do not add facts,
figures or examples that are not in my draft. List any criterion with no
supporting evidence at the end.
CRITERIA: [paste]
DRAFT: [paste]Two cautions. Some grant-makers now ask applicants to say whether they used AI, so read each funder's guidance before you submit and answer honestly if asked. And anything the centre promises to deliver (numbers served, hours opened, outcomes) is a commitment a trustee or the manager signs, not a sentence the assistant is allowed to improve. The tutorials on writing grant applications with AI and on whether funders accept AI-assisted bids go further.
Jobs that stay with a person, with the reason for each
- Safeguarding concerns. A volunteer mentions that a child seemed frightened after youth club. That goes to your designated safeguarding lead under your policy, and the facts are recorded in the format your policy requires. Do not paste it into any assistant. The material is about a named person, could harm them, and may need to be preserved exactly.
- Complaints. An assistant can help tidy an acknowledgement after the person handling the complaint has decided what to say. The investigation, the decision and the reply follow your complaints procedure, so if you do not have one, write it before you need it.
- Accident and incident records. Write what happened, in your own words, at the time. Do not ask an assistant to tidy it afterwards; a record that has been paraphrased can lose the detail that matters.
- Fire and safety checks. AI can draft a checklist from your own policy. A competent person must still walk the building, test what needs testing and sign the record.
- Room clashes and allocation. When two groups want the same slot, the decision is about fairness and your priorities. Apply your allocation policy and tell both groups why.
- Volunteer welfare and discipline. Conversations about someone's conduct, health or circumstances happen face to face or by phone, with a note kept by the manager.
Costs, accounts and a one-page policy for a small team
The free plans of ChatGPT and Claude are enough for the drafting jobs above, which are small and infrequent. If you want a shared workspace, OpenAI's help page says eligible nonprofits can get ChatGPT Business Standard seats at $8 per user a month billed annually or $10 monthly, with Premium seats excluded, and Anthropic's nonprofits page lists a Team plan at $8 a user a month with a two-seat minimum. Eligibility is verified by Goodstack, and your centre's legal form decides whether it qualifies, so check before you budget. For three users the ChatGPT route is $288 a year on annual billing ($24 a month); two seats would be $192.
The one-page policy needs five lines. Who may use AI: named staff and trained volunteers, each with their own login, since the terms of both providers forbid sharing credentials. What must never be pasted: names with health, safeguarding or behaviour information, children's photos, volunteers' personal details, and financial account details. Who approves drafts: a named person per job. What is logged: any mistake the assistant made, and the rule you added. Who reviews the policy: the committee, once a year. The tutorial on writing an AI policy for a small charity expands each line, and the trustee risk register helps the committee own it.
Start with one job: a four-week pilot for the committee
Pick booking replies, because the data is yours and the risk is low. In week one, collect your last 20 enquiries and the replies you sent, remove surnames and phone numbers, and run the prompt on all 20. Compare each draft with what you sent and mark it a match, an improvement or a problem, where a problem is a promise, a wrong price or a wrong date. In week two, fix the prompt until the 20 show no problems, then use drafts on live enquiries with the clerk approving each one. In week three, log every "For the manager" line; you will find gaps in your hire terms, and fixing those helps humans as much as the assistant. In week four, report to the committee with three numbers: minutes per reply, problems found and gaps in the terms. Then decide whether to add rota messages or the newsletter. Doing one job properly teaches the team where the assistant fails before it touches anything larger.
Three checks after a quarter
- First reply time to hire enquiries. Read it from your inbox. If it has fallen without any wrong promises, the booking drafts are working.
- Empty shifts and last-minute gaps. If rota gaps are rarer, the grid and reminders are doing their job. If they are not, the data is stale.
- The mistake log. Every error the assistant made, with the rule you added. A short log with few repeats means the rules are settling. A long one means the job needs tighter rules or goes back to a person.
Put the results in front of the committee each quarter in a short note, so the decision to continue, widen or stop belongs to the people who answer for the centre.
What community-centre managers ask about AI
Can a community centre get discounted AI tools?
Sometimes. OpenAI lists ChatGPT Business Standard seats for eligible nonprofits at $8 per user a month billed annually ($10 monthly), and Anthropic lists Claude Team for nonprofits at $8 a user a month with a two-seat minimum. Eligibility is checked by a third party, Goodstack, and depends on your legal form, so check before you budget.
Should volunteers use the centre's AI account?
Only if each has their own login and has read your one-page policy. OpenAI's and Anthropic's terms forbid sharing credentials, and a shared login also leaves no record of who pasted what. Most volunteers do not need an account at all. A manager can run the drafting jobs and share the approved text.
Can AI write a whole grant application?
It can draft sections from facts you supply, tighten wording and check a draft against the funder's criteria. It cannot know your numbers, and every figure must trace to your records. Some funders now ask applicants to disclose AI use, so read each funder's guidance first. A person signs what the centre promises to deliver.
Do we need an AI policy for a small centre?
A one-page policy is enough: who may use AI, what must never be pasted in, who approves drafts, and where mistakes are logged. It protects volunteers as well as the centre, because it tells them where the line is before they cross it. Trustees or the committee should agree and review it each year.
Further reads
- ChatGPT for Nonprofits: Volunteer Recruitment Posts and Rota Messages — Volunteer recruitment posts and rota messages, with prompts.
- Volunteer Onboarding Checklist for Charities: Build One With AI — Build an onboarding checklist for new volunteers.
- Charity Complaints Policy Template: Draft One With AI and Check It — Draft a complaints policy and what to verify in it.
- Church GDPR and ChatGPT: What Member Data Never Goes In — Member-data rules that apply to any small membership organisation.
- OpenAI for Nonprofits: Who Qualifies and What to Check First — Who qualifies for OpenAI's nonprofit pricing, and what to check first.
- AI Tool Discounts for Charities and Non-Profits: What to Claim — Which AI tools discount for charities and how to claim.
- AI Implementation for Nonprofits: A Phased Plan for a Small Team — A phased plan for a small team adopting AI.
- Venue Booking Software: Calendar, Deposits and AI Replies Compared — How to compare venue booking software on the three things roundups skip: calendar safeguards, deposit rules and AI-drafted replies, with dated prices.
- Sports Club Treasurer's Report: A Plain Summary of the Books With AI — How a volunteer treasurer builds a one-page table, has AI draft a plain summary, three talking points and member answers, then checks every figure.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: OpenAI's help page on its nonprofit programme, Anthropic's nonprofits page, SignUpGenius pricing page, Mailchimp and MailerLite free-plan pages, checked 29 October 2026. The centre, its numbers and sample outputs are illustrative.