Can AI Write a Nursery Self-Evaluation Form From Your Evidence?

Yes, as long as the evidence comes first. Keep an anonymised evidence log, then ask AI to draft each entry in a Statement, Evidence, Develop layout, using only that log and naming the impact on children. The manager reads every line, owns every judgement, and removes anything the log does not support. AI shapes the writing; it cannot know your nursery.

One premise first: Ofsted does not require a self-evaluation form. Its early years inspection information for use from September 2026 says it does not expect a formal self-evaluation and that leaders are not disadvantaged without a written one. Inspectors still ask about your strengths and areas for development in the planning call, using the inspection toolkit as the common reference, so a short, evidenced self-evaluation is the best preparation for that conversation. If a different inspectorate covers your setting, check its own guidance, and say at the top of your document that it is for the nursery's own use.

Follow me on Instagram@sagnikteaches

What a good self-evaluation is for

People often treat a self-evaluation as an exam answer. It works better as a short, honest map: where the nursery is strong, where it is not yet, and what happens next. A reader, whether a new room leader, an owner or an inspector, should see in minutes what you believe and what backs that belief.

Connect on LinkedInSagnik Bhattacharya

AI helps with the part that takes time, which is turning scattered notes into clear sentences. It does not help with the part that matters, which is deciding what is true. A model has never seen your baby room at 8am, so it cannot tell whether your statement on key-person relationships is honest. That is why the structure below puts the manager's judgement in the middle and AI at the edges.

Subscribe on YouTube@sagnikteaches

The Statement, Evidence, Develop layout

Each entry has three parts, and each part has a different author.

Notice the last column. The AI never writes the first column on its own. It can offer a cleaner phrase for something you have decided, and it can point out when a statement is vague, but the judgement remains with a person who has watched the children.

Build the evidence log first

Everything depends on this log. Without it, AI will produce plausible text with nothing behind it. A log is a spreadsheet with one row per piece of evidence and these columns: date, area (for example "communication and language"), what happened, source (where it is filed), impact on children, and who recorded it. No child names; use "a two-year-old", "three children in the pre-school room" or a group code.

Say a private nursery with 42 places adds these illustrative rows across a month:

Every impact column reads as an observable change or a count, not as an adjective. That is the habit you are building. A row that says "improved communication" with no count is a note for the team to firm up, and the AI should be told not to treat it as evidence.

If your nursery software, such as the packages compared in this nursery software comparison, can export counts (observations per room, parent logins, key-person groups), use those exports as evidence sources. Do not paste the observations themselves.

Drafting an entry from the log

Give the AI only the log rows for one area and a strict layout. This prompt works for a small nursery.

You are helping a nursery manager draft one section of an annual
self-evaluation. Use ONLY the evidence rows below. Do not add facts,
numbers, outcomes or events that are not in them. Do not give a grade
or use inspection grade words such as "strong standard" or "exceptional".

For the area "[area name]" write:
STATEMENT: two sentences on what we do and what difference it makes
to children, using the language of the evidence.
EVIDENCE: a bullet list; each bullet one row, with date and source.
GAPS: anything the statement claims that the evidence does not yet
prove. Be blunt.
DEVELOP: up to three actions, each with an owner role, a date, and
how we will know it worked. Mark anything I must decide with [MANAGER].

EVIDENCE ROWS:
[paste the rows for this area, anonymised]

The GAPS line is the most valuable instruction. Models tend to write confident statements, and asking for gaps flips them into a critic. Here is an illustrative output for the language row above:

STATEMENT: We use story sacks and repeated rereading in the toddler room to build children's interest in books and early speech. Eight of the eleven toddlers now ask for a favourite book without being prompted, according to the room leader's count.

GAPS: The count is one adult's tally for one room and one month. No baseline is given for before the story sacks began. No evidence for babies or pre-school children is supplied.

DEVELOP: 1. Record a baseline and a three-month count for toddlers (room leader, by 31 January, a repeat count shows the trend). 2. Extend the idea to the pre-school room [MANAGER: decide whether this is a priority].

This is exactly the sort of output you can work with. The statement is supported, the gap is honest, and the actions are specific. What you do next is judgement: is the baseline realistic, is the pre-school room a priority, does the nursery really believe language is a strength?

The manager's pass: what you change every time

Go through each entry with a pen and ask five questions.

1. Is the Statement true? If a team member read it, would they nod, or say "that is not quite right"? Rewrite until they nod.
2. Does each evidence item exist? Open the source. If the minutes or the register cannot be found in two minutes, the item is not evidence yet.
3. Is impact stated for children, not for the nursery? "We bought story sacks" is an action. "Eight of eleven toddlers ask for a book" is an impact.
4. Is the Develop action small enough to happen? "Improve outdoor learning" fails; "Add a mud kitchen to the garden and observe 5 children over 2 weeks" passes.
5. Would you be comfortable saying this to a parent? If not, change it.

The most common correction is tone. Models write like brochures. Strip words like "vibrant", "nurturing" and "rich" unless a specific piece of evidence earns them. A reader who has to dig through adjectives will trust the rest less.

Keeping children's data out of it

Self-evaluation needs patterns, not children. The log should never hold a name, a photo or an observation that identifies a child. If an anonymised example helps ("a child new to English who began joining in songs after three weeks"), make it too general to point to a person, and keep the full record in your nursery software, not in the AI tool.

Business plans of ChatGPT and Claude, and Gemini in Google Workspace, do not train on your content by default, and consumer ChatGPT and Claude plans let you switch training off in privacy settings. Even so, only give the AI what you could comfortably show a parent. For the wider picture on what to tell your team, see whether it is safe to use AI with children's data in a nursery and the one-hour staff briefing in nursery staff training on children's data. A short written policy helps too: the early years AI policy template sets out what is allowed.

A worked pass through one area, with numbers

Consider the "parent partnership" area for a 42-place nursery. The manager collects four rows over a term: the weekly home-learning slip in the baby room (7 of 9 parents tried an idea), a stay-and-play afternoon attended by 14 families out of 38 invited, a parent survey with 21 replies of 38 (55%), and the introduction of a daily photo message in the toddler room, which 9 of 12 families said they valued.

The AI draft states that engagement is strong. The manager's pass tightens it. Attendance at stay-and-play was 14 of 38, which is 37%, so the Statement now reads that "over a third of families came". The survey response rate is 55%, which is worth noting because 45% did not answer. The GAPS line the AI produced adds that no evidence covers families of pre-school children. The Develop action becomes: "Offer two evening slots for stay-and-play in the spring term, aiming for at least 50% of invited families (deputy, by 28 February), and extend the photo message to pre-school (room leader, by 31 January)."

Look at what happened. The AI did not discover anything. It tidied, listed gaps and offered actions, and the manager supplied the percentages and the decisions. That is the right division of labour.

A yearly review without a panic in the final week

Run the whole thing on a calendar so it does not become a rush. A workable rhythm for a small nursery looks like this.

• Weekly, 10 minutes: each room leader adds anything notable to the log. No child names.
• Monthly, 30 minutes: the manager scans the log, fixes unclear rows and asks staff for missing sources.
• Each term, 90 minutes: run the prompt for each area, review the draft entries with the deputy, update the Develop actions and tick off the finished ones.
• Yearly, one team meeting: read the full form together. Ask the team which Statements they disagree with; their answer is often the most useful part.

Share the form with the team, not only the owner. Staff who helped write an action are far more likely to carry it out. If you use a shareable Project on a business plan of ChatGPT or Claude, you can save the prompt and the layout so a deputy can run it. Do not build it as a custom GPT: OpenAI has scheduled custom GPTs to retire on 11 December 2026.

Grading the strength of each piece of evidence

Not all evidence weighs the same, and a self-evaluation that treats a single remark and a termly count alike misleads the reader. Add one column to the log, "strength", with three values you define once. Something like this works.

Then tell the AI: "Describe how strong the overall evidence is for the Statement, using the strength column. If most rows are weak, say so." A Statement that rests on three weak rows should be softened, or the Develop section should begin with "collect stronger evidence". This is more useful than a polished paragraph, and it is the kind of self-criticism a model is good at when you ask for it.

Before and after: one weak statement made honest

Here is a realistic first draft that a manager might write, or that an unguided AI might produce.

Before: "Our outstanding key-person system ensures every child feels safe, secure and valued, and all families are fully engaged in their child's learning."

Three things are wrong. "Outstanding" reads as a grade claim, and it is not even on the current inspection scale; "every" and "all" are claims no log of a dozen rows can support, and nothing shows how the system works. Apply the layout with real log rows (illustrative): key-person groups of no more than 6 children; a settling-in record for 11 new starters showing 9 settled within two weeks; a parent survey with 21 replies.

After: "Each child has a named key person, with groups of no more than six. In the autumn term, 9 of 11 new starters were recorded as settled within two weeks (settling-in records). In the parent survey, 21 families replied; 18 said they knew who their child's key person was. We want the two children still unsettled after two weeks, and the three families who were unsure, to be our focus next term."

Everything is countable, and the last sentence names a gap. The "after" version is longer and less impressive, which is the point: a reader trusts it more.

Different settings, different evidence

The same method suits other early years settings, with different logs. A childminder working alone has a shorter log, perhaps ten rows a term, and might add a column for the age band of the children. A school-based nursery class will draw on assessment points and transition visits already in the school's records. A chain with several sites can use the same layout in each and compare Develop actions across the group, which helps a regional manager spot a shared weakness.

If you are a small setting, resist the temptation to copy a template with 40 headings. A form with six areas you can evidence beats one with forty you cannot. Pick the areas that matter to your children, for example communication and language, personal and social development, parent partnership, safeguarding practice, staff development and the learning environment, and add others only when you have the evidence.

Lining your areas up with the inspection toolkit

The early years inspection toolkit grades separate evaluation areas rather than one overall judgement: inclusion; curriculum and teaching; achievement; behaviour, attitudes and establishing routines; children's welfare and wellbeing; and leadership and governance, with safeguarding judged as met or not met. Each graded area uses a five-point scale that runs from urgent improvement through needs attention, expected standard and strong standard to exceptional, and inspectors take the expected standard as their starting point.

You do not have to copy that structure, and the guidance says leaders are not expected to produce written evidence for each standard. Even so, tagging each log row with the toolkit area it speaks to pays off twice. Before the planning call, you can sort the log by area and see at a glance where your evidence is thin. And the Statements you have written become the notes you speak from when the inspector asks what you see as your strengths and priorities.

A useful prompt line for this: "For each evidence row, suggest which toolkit evaluation area it most relates to, and list any area with fewer than three rows." Check the tags yourself. A row about key-person settling-in could sit under welfare and wellbeing or under behaviour and routines, and the choice is yours. Never ask the model which grade an area would get. It has not seen your rooms, and a guessed "strong standard" written into your document can only mislead the team.

Quick checks that the form is real

Before anyone outside the nursery sees the form, spend 30 minutes on these four checks.

1. The random bullet test. Pick five evidence bullets at random and open each source. If one is missing, find out why.
2. The staff read-through. Ask a room leader to read their own area and mark any line that is not true. Treat their marks as corrections.
3. The word test. Search the document for "always", "all", "every", "never", "outstanding" and "exceptional". Reduce each claim to what you can count.
4. The year-on-year test. Compare this year's Develop actions with last year's. If the same action appears twice with no progress, say why in plain words.

Spend a few minutes on that fourth check in particular. Repeated unfinished actions are more telling than any Statement, and an honest sentence about why something stalled shows more insight than a tidy list of new plans.

Mistakes that show up in the first draft

The invented evidence. A model asked to write a "strong" statement will sometimes add a training day or a parent survey that never took place. The fix is the instruction to use only the rows, plus a check that every bullet maps to a row you can point to.

The inflated claim. "All children make excellent progress" cannot be backed by a log of 12 rows. Reduce the claim to what the rows show.

The sameness across areas. If every Statement begins "We are committed to", the document reads as filler. Ask the model to vary sentence openings, then rewrite the first line of each by hand.

The never-ending Develop list. Fifteen actions is a wish list. Limit the whole form to five or six priorities for the year, and move the rest to a "later" tab.

What goes in the finished document

A tidy final form is short. A page summarising where the nursery is strongest and where it is working on improvement, then one page per area with the three parts, then an appendix listing evidence sources and where to find them. If you use it with parents, create a one-page plain version, drafted separately, which leaves out staffing detail and anything that identifies a child.

Finally, put a date and the manager's name on it. A self-evaluation belongs to a person, and a document that says so is harder to dismiss as paperwork. Used well, the log and the form become part of how the nursery runs, and the drafting is the smallest part of the work.

Questions managers ask about self-evaluation and AI

Does Ofsted require a nursery self-evaluation form?

No. Ofsted's early years inspection information for use from September 2026 says it does not expect providers to have completed any formal self-evaluation, and leaders are not disadvantaged without a written one. Inspectors still ask about your strengths and priorities in the planning call, so many settings keep a short, evidenced version. Other inspectorates set their own expectations.

Can I put children's names into the AI?

No. Evidence for self-evaluation works with counts, themes and anonymised examples: "a toddler in the baby room", not a name. Keep observations and learning journeys in your nursery software and give the AI only the summary lines you have written.

How often should the form be updated?

Treat it as a living document. Update the evidence log weekly or monthly, review the Statements each term, and do a full yearly review with the team. A form updated only before an inspection reads as written for the inspection.

Can AI tell us which inspection grade we would get?

It should not. Inspectors now grade each evaluation area on a five-point scale, from urgent improvement to exceptional, against the toolkit's grading standards and what they see in the setting. Ask AI to check that each Statement is backed by evidence and which area it relates to, never to predict a grade.

Further reads

• Nursery Fees Policy: Deposits, Late Payment and Holidays With AI — Another policy a manager can draft with AI and own.
• Nursery Data Protection Policy: Draft and Update It With AI — Keep your handling of children's data in order.
• Choosing Nursery Software With AI: What to Compare — What to compare when choosing nursery software.
• How to Increase Nursery Occupancy: Open Days, Reviews and AI — Use your self-evaluation themes in open days.
• AI Chatbots for Nursery Enquiries and Visit Bookings — Handle enquiries while the team gets on with evidence.
• Preschool Newsletter: A Monthly Issue From Your Notes, With AI — Share impact stories with parents, with consent.
• AI Tools for Preschool Owners: Parent Updates and Daily Reports — Draft daily reports and parent updates with AI while children's names, photos and health details never reach the tool.
• Care Inspectorate Self-Evaluation: Draft the Evidence Summary With AI — A Scotland-specific worked self-evaluation: quality framework headings, an evidence summary drafted with AI prompts, and a provider who owns every judgement.
• 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.
• Nursery Late Collection Policy: Wording, Fees and Parent Messages With AI — A late collection policy a nursery can actually enforce: grace period, fee wording, repeat-lateness steps, emergencies and the messages to send parents.
• Nursery Ofsted Inspection Checklist: An Evidence Folder With AI — What a nursery should have ready for an Ofsted visit, how to build an evidence map and document index with AI, and how to practise staff questions.
• Nursery Welcome Pack: Letters and Settling-In Guide With AI — What goes in a nursery welcome pack, with prompts for plain-English letters, a settling-in guide, a key person introduction, forms and checked translations.
• Preschool Parent Handbook: Draft It With AI and Check State Rules — Use AI to outline the handbook and rewrite policies in plain English, then check every section against your licensing authority's own published rules.
• AI for Nurseries: Drafting Risk Assessments Managers Can Sign Off — How a childcare nursery manager can use AI to draft room and outing risk assessments, then walk the space, verify the rules and sign off.
• Can a Nursery Use AI for Accident Reports? What to Keep Out — A facts-only workflow for turning keyword notes into accident report wording: a child code rule, parent message wording and a sign-off and reporting check.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Ofsted, Early years inspection information and Early years inspection operating guide for inspectors, both for use from September 2026 (gov.uk, updated 28 August 2026); OpenAI custom GPT retirement FAQ; OpenAI and Anthropic data-control pages (all checked 4 November 2026). Inspection guidance changes, so check the current version before relying on this tutorial.