A daycare can use ChatGPT for work that contains no children's details: parent newsletters, policy drafts, rota patterns, job adverts and generic messages with placeholders. It should never be given names, photos, health, behaviour, allergy, safeguarding or custody information. Free and business plans differ on training and admin controls, but neither changes the keep-out list.
The useful way to think about it is not "is ChatGPT allowed?" but "does this prompt contain a child?" If the answer is no, you are in the same position as any small business drafting copy. If the answer is yes, stop and write that part by hand or in your childcare software. Most slips are not careless: they happen when someone pastes a real incident note or a parent's email to "tidy the wording", so the habit to train is placeholders, not a blanket ban.
The test every prompt has to pass
Before anything is pasted, a staff member asks three questions:
- Could a stranger identify a child from this text? A name is the obvious route, but "the two-year-old in Sunflower room who bit twice on Tuesday" can also identify a child to anyone who knows the centre.
- Does it describe health, behaviour, family circumstances or safeguarding? Even without a name, these are the categories parents and licensing bodies care most about.
- Would I be comfortable if the parent read this prompt over my shoulder? A gut check, and a surprisingly good one.
If any answer is yes, the task belongs in your childcare management software or on paper. Tools built for early years, such as the platforms compared in our Brightwheel and Procare comparison, have their own contracts and controls for children's records. A general chatbot does not.
Jobs ChatGPT can safely do for a daycare
Here are the uses that pass the test, grouped by how often they come up. Each one works with no child information at all.
Parent communication in general terms
Closure notices, fee reminders, event invitations, weather and illness policy reminders, and "what to pack" messages are the same for every family. Say a centre has to close for a staff training day. The prompt:
Write a short, warm message to parents of a childcare centre. We are closed on [date] for a staff training day. Fees are not charged for that day under our contract. Normal hours resume [date]. Tone: friendly, plain English, under 90 words. Do not mention any child by name.A typical draft (illustrative) comes back with a greeting, the date, a sentence about fees, a thank-you and a sign-off. What you fix: check the fee line against your actual contract, add your name and the contact route, and cut any sentence that promises something you have not decided, such as "we will offer a make-up day".
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowNewsletters and themed activity ideas
A monthly newsletter is a good fit because its raw material is your plan, not your children. Give it the theme ("autumn: leaves, conkers, mud kitchens"), the dates and any reminders, and ask for a draft with three short sections. Add real photos and any mention of a child's achievement yourself, inside your own newsletter tool, with the consents you already hold.
Policies, procedures and checklists
ChatGPT is good at first drafts of a sun-cream policy, a late-collection procedure, a staff mobile-phone rule or a risk assessment template. It does not know your licensing requirements, so the draft needs a line-by-line check by the person who holds that responsibility. A useful trick is to paste your licensing body's published requirement text into the prompt (public text, so nothing sensitive) and ask the draft to follow it, then check it anyway.
Rotas, cover plans and shift swaps
Staff names and hours are personal data about adults, so keep it light: use roles or initials. Describe the pattern ("three rooms, open 07:30 to 18:00, ratios of one adult to four under-twos, a 30-minute break for each of eight staff") and ask for a rota grid or a list of coverage gaps. Check ratios yourself. The model can produce a tidy table that breaks your ratio at lunchtime.
Recruitment and staff communication
Job adverts, interview question banks, induction checklists and thank-you emails to applicants have nothing to do with children. Use real applicant details only in your own systems, not in the chat.
What to keep out, written as a list for the staff room
A never-paste list is only useful if it is specific. This one can go on the wall:
| Never paste | Why it catches people out |
|---|---|
| Child names, nicknames, dates of birth, photos | "Just the first name" still identifies a child to families and staff |
| Allergies, medication, medical or additional needs | Special-category information; a wrong paraphrase can also cause harm |
| Behaviour notes, incident details, observations | Often feels like "just tidying the wording", which is the trap |
| Safeguarding concerns and conversations with outside agencies | Confidential and may be legally sensitive |
| Parent names, contact details, custody or pickup arrangements | Directly personal and often contested |
| Learning journals and progress reports with real content | A full profile of a child in one pasted block |
| CCTV stills, registers and sign-in sheets | Named children plus times and presence |
| Staff disciplinary or health matters | Employment confidentiality |
The pattern is that the dangerous prompts are rarely malicious. They are "tidy this incident note" or "make this parent email sound kinder" with the real text pasted in. Teach the placeholder habit instead: replace every real detail before the paste, and put the details back afterwards in your own system. Our one-hour staff briefing includes a drill for exactly this.
Turning a real garden fall into a safe request
Suppose a toddler room leader has to write a note to a parent after a minor fall in the garden. The unsafe version of the prompt pastes the child's name, the room, the injury and the time. The safe version works like this.
Step 1, strip it. The leader writes, in their own notes: "[Child] tripped on the path at 10:20, grazed left knee, cleaned with water, plaster applied, comforted, ate snack normally afterwards."
Step 2, ask for the tone, not the facts. The prompt: "Write a calm, reassuring message from a childcare room leader to a parent. A child had a minor trip outside this morning with a small knee graze. It was cleaned and a plaster was applied, and the child is playing as normal. Keep it under 80 words and sound human. Use the placeholders [parent] and [child]."
Step 3, finish by hand. The leader pastes the draft into the childcare app, swaps in real names, checks the facts against the accident record and sends.
The AI never saw a child, a time or a name. It supplied only wording. If the incident were more serious, the message would not be drafted this way at all: serious injuries follow your incident procedure and licensing notification rules, not a chatbot. Our tutorial on factual, name-free incident narratives covers the careful version of that job.
Free versus business plans: what actually changes
The plan matters for settings and control, not for the keep-out list. The facts as checked on OpenAI's own pricing and help pages in November 2026:
| Free or personal plan | Business workspace | |
|---|---|---|
| Training on your chats | On by default on Free, Plus and Pro; you can switch it off in data controls ("Improve the model for everyone") | Not used for training by default |
| Who controls settings | Each person for their own account | An admin sets rules for the workspace |
| Shared prompts and files | Projects can be shared on every plan | Shared Projects with workspace members, plus admin controls |
| Cost | Free tier; Go is $8 and Plus $20 a month at list price (Pro costs more) | From two users; $20 a user a month billed annually or $25 billed monthly at list price |
| Fit for children's records | No | Still no; use childcare software for that |
For a single-site daycare with a manager and a deputy, the sensible option is usually a two-seat business workspace for the leadership team, with a shared Project holding your approved templates, tone notes and public policy text. That gives you shared settings and an off-by-default training position in one place. For a centre that just wants to try it, a free account used only for non-child tasks costs nothing, but then each person manages their own privacy toggles. If a staff member uses ChatGPT's free plan, tell them to check data controls the first day. For a one-off generic job on a personal account, Temporary Chat is a sensible extra: OpenAI says those chats stay out of history, do not create memories and are not used to improve its models, though they may be kept for up to 30 days for safety purposes. It is still not a place for a child's details. OpenAI changes plan features regularly, so confirm the current list on its pricing page before you buy.
One more point about older advice: OpenAI retires custom GPTs on 11 December 2026 and is moving their creators to plugins, so do not build your daycare tools around them. A shared Project in ChatGPT, or an equivalent in another assistant, does the same job of holding instructions and reference files.
COPPA, licensing and where the lines sit
Children's privacy rules vary a great deal by place, so this section stays general. COPPA, the Children's Online Privacy Protection Act, places requirements on operators of websites and online services directed to children under 13, and on other operators that know they are collecting personal information online from a child under 13. A daycare typing a child's details into an adult chatbot is not what COPPA was mainly written for, but that does not make it safe: it still creates a data-protection and trust problem with parents, and it may breach your licensing conditions or your own parent contract.
Licensing bodies and local regulators often set separate rules on record keeping, confidentiality and who may access children's files. The practical way to handle that uncertainty is to ask the body that licenses you two questions in writing: "Do you have any guidance on staff using generative AI tools?" and "Does that guidance affect records or communications about children?" File the reply. Whatever the answer, the cautious position is the same: no child-identifiable information in general-purpose AI tools.
Wherever you operate, look for the equivalent children's and general data-protection rules. Our tutorial on children's data and AI in a nursery goes through those questions in more depth.
Five prompts a daycare can use this week
These all pass the test. Replace the square brackets with your own details.
1. The monthly newsletter skeleton
You are the editor of a childcare centre newsletter. Theme this month: [theme]. Dates to include: [dates]. Reminders: [reminders]. Write a 300-word newsletter with a friendly opening, three short sections and a closing line. No children's names. Plain English, British spelling.Expect a decent structure with slightly generic phrasing. Fix by adding one genuine local detail from your week and removing any promise (such as "a special visitor") you have not arranged.
2. A policy first draft with a checking list
Draft a [late collection] policy for a childcare centre. Include: purpose, what staff do at 5, 15 and 30 minutes late, how charges are explained to parents, and review date. Then list five questions the manager should check against our licensing requirements before approving it.The five checking questions are the valuable part: they force the review that people skip.
3. The rota gap finder
Our centre has 3 rooms, open 07:30 to 18:00. Required adult-to-child ratio: [ratio]. Staff roles and availability (roles only): [list]. Find the time slots where cover is thin and suggest two ways to fix each.Check the sums yourself. A model will sometimes count a staff member on break as present.
4. Tone adjuster for a sensitive general message
Rewrite this notice so it is clear and kind, and under 100 words. It is about a fee increase from [date]. Facts: [new fee], reason: [reason], notice period: [period]. Do not apologise more than once.5. An interview question bank
Write ten interview questions for a room leader role in a childcare centre, covering safeguarding awareness, working with parents and handling a busy room. For each, say what a strong answer would include.All five use public or generic information. None needs a child.
Three drafts and what a careful manager changes
The quickest way to see what "check it" means is to look at output. These are illustrative drafts, with the fix a manager would make.
A fee reminder that promises too much
Suppose the prompt asks for a polite reminder about overdue fees. The draft says: "Please settle your balance by Friday to avoid any further action." The phrase "further action" is vague and sounds like a threat, and the deadline is something the model invented. The manager rewrites it to match the actual contract: "Our records show a balance of [amount] due on [date]. Please pay by [date] or contact me if you would like to talk about a payment plan." Specific, calm, and it matches what the centre will really do.
A policy paragraph that sounds right and is not
Ask for a sickness policy and a model may write: "Children must stay at home for 48 hours after any illness." That is a clean sentence and a possible mismatch with the exclusion periods your licensing body or health authority publishes, which differ by illness. The manager's job is to replace the blanket rule with a table of conditions and periods copied from the official guidance, and to say who updates it. Treat any number, period or ratio the model gives you as unverified until you have seen it on an official page.
A newsletter intro with an invented event
Asked for an autumn newsletter, a model may open with "Join us for our annual harvest festival on the last Friday of October." You have no such festival. This is the commonest failure with creative drafts: the model fills gaps with plausible events, dates and quotes. Read the draft with a pen and strike anything you did not supply.
A quick sum for a ten-person centre
Say a centre has a manager, a deputy, six room staff and two cooks. The manager spends about three hours a week on messages, newsletters and policy upkeep, and the deputy about two on rotas and recruitment. If drafting help cuts that admin time by a third, which is a cautious guess to test rather than a promise, the saving is roughly 100 minutes a week, or around 7 hours a month across the two of them. A two-seat business workspace at list price ($20 a seat billed annually, $25 monthly) is $40 to $50 a month. If the manager's time is worth even $20 an hour, the break-even is two to two and a half hours a month. Measure your own first month, because it is the pattern of use that matters, and the sums stop being worth it if staff spend the saved time rewriting poor drafts.
A first-month plan that keeps the risk low
- Week 1: Write the one-page rule and the never-paste list. Choose the account type. Set data controls.
- Week 2: The manager uses ChatGPT for two generic jobs only, such as a newsletter and a policy draft, and logs what had to be fixed.
- Week 3: Add one more person. Run the first ten-minute prompt check together.
- Week 4: Review the log. Keep the uses that saved time without extra fixing, drop the rest, and add the two best prompts to the shared Project.
Rolling out slowly also gives you a story for parents: you started small, with admin that never touches a child, and you can say exactly what the tool is used for.
Writing the staff rule in one page
A policy staff will read is short. A workable one-page version has five parts: the approved accounts (for example "the shared business workspace only"), the allowed uses, the never-paste list, the placeholder rule, and who to ask when unsure. Add a sign-off line so each staff member confirms they have read it, and revisit it every term. Our tutorial on ChatGPT memory is worth reading before you finalise it, because memory and chat history affect what could resurface later. For decisions about wider tools, such as an AI phone receptionist, see the daycare receptionist cost breakdown.
How to check it is working
After a month, run a ten-minute audit. Ask each person who uses ChatGPT to show you two recent prompts, with their permission and without anything sensitive on screen. You are looking for names, dates of birth or incident details. Count how many prompts used placeholders, how many saved time, and what was edited before sending. If you find a slip, treat it as a prompt to improve the training, not a disciplinary moment. If a child's details were entered, note the date, delete the chat, and decide with your data-protection lead whether it needs recording or reporting under your rules.
This tutorial is practical guidance, not legal advice. Check your licensing requirements and privacy obligations with the bodies that apply to you.
Daycare owners' questions about ChatGPT
Is a child's first name safe to type into ChatGPT?
Treat it as no. A first name plus a room, a date or a medical remark can identify a child. Use a placeholder such as [child] or [A] in prompts and add the real name in your own document afterwards.
Does turning off model training make it safe for child data?
No. Switching off training stops your chats being used to improve models. It does not turn a general chatbot into a system approved for children's records, and your licensing and privacy duties are unchanged. Keep child data out regardless.
Can staff use their own personal accounts?
They can, but you lose control of settings, history and who sees what. Decide as a policy whether staff use one shared business workspace, or personal accounts for non-child tasks only, and put it in writing.
Should we tell parents we use AI?
If AI only helps draft generic newsletters that staff check, a line in your privacy notice is sensible. If you ever planned to put children's information into any AI tool, you would need much stronger consent and a tool built for it.
Can ChatGPT write our policies?
It can draft a first version in minutes, but it does not know your licensing rules. Treat the draft as a starting point, then check it against your licensing body's requirements and have the manager or owner approve the final text.
Further reads
- How Nurseries Use AI to Write Daily Reports for Parents — Daily parent reports with AI, step by step.
- Is It Safe to Put Customer Data Into ChatGPT? — The general customer-data rules behind this list.
- AI Chatbots for Nursery Enquiries and Visit Bookings — Handling enquiries and visit bookings with a chatbot.
- Choosing Nursery Software With AI: What to Compare — What to compare when choosing childcare software.
- Tapestry vs Famly for a Small Nursery: Where Does AI Fit? — Where AI sits inside two nursery platforms.
- Daycare Inspection Checklist: Prepare Your Records With AI — Build a daycare inspection checklist from your licensing rules, check staff and child records against it with AI, and keep a corrective-action log.
- AI for Daycares: Nine Admin Jobs From Enquiry to Pickup Message — Enquiries, tours, waitlist, forms, rotas, parent updates, policies, job ads and invoices: nine daycare jobs AI can draft, each with its risk level.
- Child Care Subsidy Billing: Can AI Match Payments to Attendance? — How a daycare can match subsidy remittances to attendance with AI-built formulas, find shortfalls and write top-up notices without names.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: OpenAI ChatGPT and ChatGPT Business pricing pages; OpenAI help centre articles on data controls, Projects and custom GPT retirement; FTC page on the Children's Online Privacy Protection Rule. Checked November 2026.