A preschool can use AI to plan a month of social media posts safely by letting it see only the activity notes and the calendar, never the children or their photos. Check every picture against a consent register, show hands and work instead of faces where you can, and have a second person approve each post.
The temptation with a busy page is to let AI do everything, including choosing the pictures. Resist it. The privacy decisions in a setting that cares for young children belong to named adults using a written register. What AI does well here is the slow, repetitive writing: turning "paint trays out, wellies on" into five good captions, spacing a month of posts so the page does not go quiet for two weeks and then flood, and rewriting a stiff announcement so a parent actually reads it.
The photo consent register that everything else depends on
Create a register with one row per child and a column for each use. Parents tick each use separately at enrolment, not as one blanket permission. A sensible set of columns: photographs in the child's own learning record; photographs in room displays inside the building; photographs in the newsletter to enrolled families; photographs on the setting's public social media page; and photographs on the website. Add columns for the date consent was given, the date it was last reviewed, and a notes column for conditions ("hands and back-of-head only").
| Child code | Learning record | Room display | Newsletter | Social media | Notes |
|---|---|---|---|---|---|
| C014 | Yes | Yes | Yes | Hands only | Reviewed Sept |
| C015 | Yes | Yes | No | No | Parent asked for no photos outside the room |
| C016 | Yes | Yes | Yes | Yes, no name | Review at January |
Keep this register in your management system or a protected sheet, using codes rather than names where you can, and keep the key that links code to child in your usual records. The register is for people. Do not paste it into an AI tool. If you want help building the consent questions themselves, the enrolment form tutorial shows how to write them in plain English and keep them out of the way of the rest of the paperwork.
Making consent easy to change
Families change their minds, and a separated family may need a different arrangement. Write a one-paragraph process: how a parent tells you (in writing to the office), who updates the register (named person, within two working days), and what happens to posts already published (you remove them within a stated time, if the platform allows). Test it once with a made-up withdrawal: can you find every post that shows that child, and can you delete it? If the answer is "not easily", that is a reason to post fewer faces, not more.
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 nowRules for what a safe post looks like
Write the rules on one page and pin it to the staff room wall. The AI never touches this step, but your caption prompts will refer to it. A workable set, which you should adjust against your licensing authority's guidance and your own safeguarding lead's view:
- No child's full name, ever. First names only where the parent has expressly agreed, and many settings use none.
- No faces unless the register says yes for social media, and even then prefer work, hands, materials and rooms.
- No uniforms, bags or items that show a child's name, and no photo that shows a child's home route or a door code.
- No other children in the background unless they also have consent. Check the corners and reflections.
- No location tags that pinpoint a child's regular movements, and no live posting from an outing while children are there.
- Photographs are taken on a setting-owned device, stored in one place, and deleted from the camera after upload.
- Comments naming children are removed; staff do not reply with names either.
Examples of safe subjects: a close-up of a mud kitchen, finished collage on a drying rack, a row of wellies, a hand holding a magnifying glass, an empty role-play area set up for the week, a staff member reading with their back to the camera and a book cover in view. Compare the response to these with the occasional cleared-face post for a few months before deciding that faces are needed at all; the activity and the caption may carry more of the interest than you expect.
Briefing the AI so captions never contain a child
Give the assistant a standing brief that you save once, in a Project in ChatGPT or Claude, or in Gemini, where Google is replacing Gems with skills (personal-account Gems go read-only on 17 November 2026; Workspace business accounts follow in March 2027). The brief states your voice, the rules above as hard constraints, and the format of what you want back. Then each week you give it only the activity note.
You write social media captions for a small preschool.
Voice: warm, specific, calm, no exclamation marks, no emojis except one
at most per post. British English.
HARD RULES: never include any child's name or age beyond a room name;
never describe a child's face, appearance or behaviour as an individual;
never say where a child lives or how they travel; never name a staff
member without being told. Write about the activity, the materials and
what the group did, using "the children" or "the pre-school room".
Each caption 25-60 words. Give 3 versions per activity, each with a
different first line. Add alt text (under 20 words) describing the scene
without people's features. Then list any phrase you think a parent could
read as a promise or claim.
ACTIVITY NOTE: [paste the staff note]Illustrative note: "Tuesday, mud kitchen, children made 'soup' with leaves and sticks, lots of counting spoons." One version of the output: "Soup of the day: three leaves, two sticks and a lot of stirring. The mud kitchen is open again, and the counting has been very serious." Alt text: "A row of pots and spoons on a wooden mud kitchen shelf, with leaves and sticks in the pots." The assistant's list of risky phrases might flag "a lot of stirring" as harmless, but flag "very serious counting" as slightly overclaiming if nobody counted. You decide. What would you fix? Delete anything that implies learning outcomes you did not observe, and shorten any version that sounds like an advert.
A worked month for a setting with 40 children
An illustrative case. A preschool of 40 children asks every family at enrolment about photo consent for social media. 31 families say yes in some form: 14 allow faces without names, 11 allow hands and work only, and 6 allow only images with no child visible. 9 say no. That means that for a typical group photo, the chance that every child in the picture is cleared for faces is small. With only 14 of 40 cleared for faces, a picture of a table of eight children will usually contain at least one who is not. So the plan treats faces as the exception.
The manager sets a target of 12 posts for the month, three a week: two activity posts with no faces, and one post per week that is not a photo at all, such as a reminder about wellies, a question for parents, or a staff welcome with the member's consent. Writing 12 captions by hand took about three hours last term, mostly staring at a blank box. Now each Friday the room leaders add activity notes to a shared document; on Monday the manager pastes the week's notes into the saved prompt, picks the best three versions, and gives the draft to a second person for approval. The writing takes about 20 minutes a week. Over four weeks that is under an hour and a half against three hours, a saving of roughly one and a half hours a month. That is a modest figure and a fair one, and the real gain is that the posts follow the rules each time.
The approval step: two people, one short list
Make approval a habit with a script. The person who drafted the post is never the person who approves it. The approver reads each caption and looks at the image, and ticks five questions: Is every child in the image cleared in the register for this use? Are there other children, names, uniforms, door codes or location details in the picture or its corners? Does the caption name or describe any individual? Does it make a claim we cannot back up, such as "all children made progress"? Would we be comfortable if the family concerned read it?
Keep the approval record simple: a column in the posting calendar with the approver's initials and the date. If the approver is unsure, the answer is no and the post waits. A post that goes live a day late is a non-event, while a post that has to be taken down after a parent complains is a conversation you would rather not have. You can add an extra rule for posts that mention any child's achievement: they need the parent's agreement for that specific post, even where general consent exists.
A posting calendar that keeps the page steady
A simple calendar prevents the stop-start pattern. Use one row per week, with a theme and a content type, and leave one slot empty for something unplanned (a lovely moment, a weather change). Ask the assistant to build the calendar from your term plan and events list: no pictures of children go in; it needs only themes and dates.
| Week | Post 1 | Post 2 | Post 3 |
|---|---|---|---|
| 1 | Mud kitchen close-up (no faces) | Autumn tray: finished leaf prints | Reminder: wellies from Monday |
| 2 | Garden in the rain, wellies in a row | Story corner: new book on the shelf | Open day dates (graphic from template) |
| 3 | Playdough hands (consent: hands only) | A question for parents: favourite song? | Staff welcome (with consent) |
| 4 | Pumpkin patterns on display | Closure notice for training day | Thank you to families for spare clothes |
A graphic made in a design tool, with text only, is a sensible fallback for weeks when nothing photographable is happening. If you want help on the broader question of how much of a page to hand to AI, should a small business use AI to write its social posts covers it, and whether AI can run social media on autopilot explains why this is a setting that should never be automated unattended. That is a good rule here: AI drafts, humans decide and publish.
Background children, location tags and other posting slips
- The background child. A parent spots their child, who has no consent, behind a table. Check corners, reflections in windows and shelves before posting, and crop if needed.
- Metadata and tags. A photo carries location data or a tagged venue that shows where the children are at a given time. Check what your phone attaches, switch location off on the setting's device, and avoid posting an outing in real time.
- A caption that names the child through a side door. "Our youngest, who arrived last week from the baby room" identifies a child to the families who know. Describe the group, not the individual.
- AI imagery that misleads. A made-up picture of smiling children in a bright classroom that does not look like your setting. Use your own pictures or clearly illustrative graphics.
- Consent drift. A consent from two years ago for "social media" now covers a platform that did not exist. Review the register once a year, and when you start using a new platform.
- Comments and messages. Parents tag each other, or a stranger asks questions. Set who monitors the page and how fast, and limit comments where the platform allows.
A two-month audit of posts against the register
After two months, run a short audit: choose five posts at random and check each against the register. Ask a colleague who did not approve them to confirm. Review the page for any picture you would not have chosen, and count how many posts you published against your plan. If three of five fail the register check, the problem is not the AI; it is the picking of images, so slow down and move to hands and work only. Note enquiries too. Compare which posts bring messages: an open-day date or a clear "how to book a visit" post may well outperform any photograph, and you can link your posts to those, as the occupancy and open days tutorial describes.
Rewriting a stiff notice so parents read it
Not every post is an activity. Closures, fee reminders and open-day notices are where captions go stiff. Ask the assistant to rewrite them while keeping every fact. Original, illustrative: "Please be advised that the setting will be closed on Friday 13 November for staff professional development. Normal service will resume on Monday." A rewrite with the rules applied: "We are closed on Friday 13 November while the team trains, and we open as usual on Monday. See you then." Check that the date, weekday and reopening day survived, and that "trains" is true (if it is an inspection or a day off, say so plainly). If you want the assistant to build a post that is also a graphic, ask it for the text to go on the picture in under 12 words, then make the picture in your design tool from your own colours and logo.
What to do when a family withdraws consent
A worked case shows why the register has to be searchable. A parent emails to say they would like no pictures of their child on the public page. Within two working days the named person changes the register row to "No" for social media and the website, adds the date, and replies to confirm. Then comes the part many settings miss: finding past posts. Search the page for posts from the child's room around the dates the child attended, check each image, and remove or crop any that show the child. Record the number of posts checked and removed, so that you can answer if the parent asks.
The lesson is a design choice. If you post mostly hands, work and rooms, the search takes minutes, and a withdrawal costs you almost nothing. If you post group photos of faces, a withdrawal means hunting through a year of pictures. That alone is a good reason to adopt the "faces are the exception" rule before the first withdrawal arrives. Offer parents a simple list of alternatives when they withdraw, such as hands-only, learning record only, or newsletter to enrolled families only, since many who say no to a public page are happy with the private uses.
Setting up the page so mistakes are harder to make
Some protection lives in the platform's settings, not in anyone's judgement. Read the help pages of the platforms you use and set what they allow: who can comment, whether tags need approval, whether the page can be messaged, and whether posts can be shared onward. Turn off any feature that automatically adds a location. Use a shared login or a business-manager style account owned by the setting, never a member of staff's personal profile, and use two-step sign-in so that a lost phone does not give away the page. Keep a list of who has access and review it whenever a member of staff leaves, on the same day.
Finally, keep a short "if it goes wrong" note taped inside the folder with the register: who to tell first (the manager and the designated safeguarding lead), how to remove or hide a post quickly, who to inform (the family concerned, and your licensing authority if its rules require it), and how to record what happened and what changed. Practise it once with a made-up case. A plan that has been read aloud in a staff meeting is far more likely to be used at 6pm on a Friday than one that sits in a policy file.
Policy, training and where to go next
Write a half-page social media policy: who may post, what devices are used, how consent is checked, the approval step, how comments are handled and how a family withdraws consent. Brief staff once a term using the same points as the one-hour AI briefing on children's data, and treat the AI tool the same way as any other service that handles your content: choose a plan whose data settings you have read, and keep personal data out. If you would like a second pair of eyes on the process for your own setting, that is the sort of thing my AI implementation consultation covers, and you can reach me through the contact page.
Questions about children, photos and posting
Is it safe to put children's photos on a preschool page?
Only with specific, informed consent for that use, and with less than you might expect: many settings post hands, activities, empty classrooms and finished work instead of faces. Parents can withdraw consent, so you need a way to find and remove past posts.
Can AI generate pictures of children for our posts?
It can, but pictures of invented children may mislead families about your setting, so label them clearly or avoid them. For real activities, photograph the activity from an angle that shows hands, materials or the room, not faces.
Should staff post from personal phones?
It is safer not to. Use a setting-owned device and a written rule on where images are kept and when they are deleted, so no child's picture sits on a staff member's own phone or cloud account. Then check whether your licensing authority or your safeguarding guidance says anything specific about staff phones.
What if a parent comments with another child's name?
Remove or hide the comment, message the parent kindly, and set the page's comment controls to limit replies where the platform allows. Your policy should say that comments naming children are removed.
Further reads
- Is It Safe to Use AI With Children's Data in a Nursery? — Set the line on what children's data may enter an AI tool.
- Preschool Newsletter: A Monthly Issue From Your Notes, With AI — Share the same monthly notes with enrolled families by email.
- Charity Appeals and Social Posts With AI That Keep Your Voice — Another sensitive-audience posting routine, with a voice guide.
- Best AI Social Media Tools for Small Businesses (2026) — Compare scheduling tools before you pick one for the page.
- How Much Time Can AI Save on Social Media Management? — Put realistic numbers on the time AI saves in posting.
- AI for Soft Play Centres: What Party-Booking Automation Costs a Month — Price party-booking automation for a soft play centre per month, and decide which enquiry, deposit and form jobs it should take over.
- Dance Studio Handbook: Parent Policies Drafted With AI — Draft a dance studio parent handbook with AI: a decision sheet, sections for attendance, costumes and behaviour, and a promise audit before sign-off.
- Dance Studio Registration: Enrolment Emails and Waitlists With AI — A dance studio's registration form, seven enrolment emails and three waitlist messages, drafted with AI and checked against consent and data-minimisation rules.
- Sports Club Code of Conduct: Player and Parent Versions With AI — Player and parent versions, a five-step sanctions ladder and a safeguarding link, drafted with AI and approved by your committee.
- Can AI Draft a Martial Arts School Waiver for Children? — A children's martial arts waiver drafted with AI in half an hour, then the consent, injury, photo and data checks a solicitor must make.
- Can AI Plan a Dance Recital? Schedules, Costumes and Parent Emails — How a dance studio can use AI for a 120-dancer recital: running order with clash check, rehearsal blocks, costume sizing, backstage rota and parent emails.
- After School Club Registration Form: Consents Built With AI — A field-by-field club registration workflow with separate permissions, authorised collector details and a chase sequence for missing forms.
- 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.
- How to Start a Preschool: First-Year Costs and Where AI Fits — A first-year cost table for a small preschool, the licensing route, enrolment targets, and the forms and parent messages AI can draft from day one.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: OpenAI's Projects help page, Anthropic's help page on Claude projects and Google's Gemini help page on the move from Gems to skills, checked November 2026; your licensing authority's safeguarding and photography guidance; the comment and tagging settings of each social platform you use; and the consents your own families have signed.