# Should Teachers Use AI to Write School Reports? Where to Draw the Line

Canonical URL: https://blog.theaieducator.io/posts/should-ai-write-school-reports
Publication: Dan Fitzpatrick Insights
Author: Dan Fitzpatrick
Topic: AI Adoption and Organisational Change
Published: 2026-10-05T07:25:00.000Z
Modified: 2026-10-05T07:25:17.182Z

Thirty-nine percent of teachers already use AI on pupil reports, and most schools have no rule for it. AI may write the sentences. It may not supply the particulars. Here is the test, and the six lines to settle before this term's reports go home.

## In brief

Teachers should use AI on the prose of a school report and never on its particulars, and the school should put that in writing before reports are drafted. In a Bett survey of 1,033 teachers conducted by YouGov in June 2026, 39 percent said they already use AI to write pupil reports, against 8 percent who use it to mark work. Dan Fitzpatrick's Three Sources Test asks where each sentence came from: the teacher's own observation of that child, the school's own records, or the model, which has never met the child and will supply a plausible particular to fill any gap the first two leave. Sentences from the first two sources may be drafted, shortened or translated by AI. Sentences from the third may not be sent. The decision a leader has to make is therefore not whether staff may use AI on reports, but what a teacher must supply before the tool runs.

## Key takeaways

- AI may write the sentences of a school report but may not supply its particulars, and the school should say so in writing before reports are drafted.
- In a Bett survey of 1,033 teachers carried out by YouGov in June 2026 and reported by Schools Week on August 27, 2026, 39 percent of teachers said they use AI to write pupil reports and 39 percent to write letters to parents, against 8 percent who use it to mark work.
- In England the head teacher, not the class teacher, prepares the annual report under the Education (Pupil Information) (England) Regulations 2005, and Schedule 1 requires brief particulars of achievements, which is why the rule belongs to leadership.
- Requiring staff to check the output is the wrong line, because a teacher cannot reliably detect a sentence that is merely plausible about a child she teaches, and a parent cannot detect one at all.
- The Three Sources Test asks which of three sources each sentence came from: the teacher's own observation, the school's own records, or the model. Only the first two may go home.
- Teacher Tapp found in June 2025 that 27 percent of head teachers believed their school allowed AI with specific rules while only 11 percent of classroom teachers said those rules existed, so most schools that think they have settled this have settled it only upstairs.
- A lighter rule is defensible where the report was never carrying a particular, such as a grade and an effort code drawn from a departmental bank, or a translation of a report already written and signed.
- If a teacher cannot name three particulars for a child, the model should not be invited to guess: that gap is information about a child nobody has looked at closely enough.

Yes, with one line held: AI may write the sentences, but it may not supply the particulars. A teacher can hand a model three specific things she noticed this term and ask for readable prose. She cannot hand it a name and a grade and ask it to fill in a child. The first is drafting. The second is invention, and it goes home to a parent with the school's name on it.

That line is not where most schools have drawn it. Most have not drawn one. In a Bett survey of 1,033 teachers carried out in June 2026 and [reported by Schools Week](https://schoolsweek.co.uk/ai-use-cuts-teacher-workload-but-not-hours-worked-study-suggests/) on August 27, 2026, 39 percent of teachers said they use AI to write pupil reports and 39 percent to write letters and emails to parents, against 8 percent who use it to mark work. Teachers have already picked the job they want AI to take, and it is not marking. It is the thing that goes to the family. In the same survey, 41 percent said their school had given them no training or support on AI at all.

So the question this term is not whether AI is writing reports in your school. It is what a teacher has to put in before the tool is allowed to draft.

## Should teachers use AI to write school reports?

Teachers should use AI on the prose of a report and never on its particulars, and the school should say so in writing before reports are drafted.

That is not a compromise between enthusiasts and skeptics. It follows from what a report is. A report is the one document a school produces whose entire value sits in the specifics: this child, named, did this thing, in this subject, this term. Strip the specifics and nothing is left but a grade and the shape of a sentence. Specifics are what a language model cannot have about a child it has never met, and what it will produce on request anyway, because producing plausible text is what it is for.

Ask a model for a warm paragraph about Ayesha in Year 5 science, give it only a grade, and it will not refuse. It will tell you she has grown in confidence in group work and contributes thoughtfully to discussion. Those sentences may even be true. Nobody in the building knows, and that is the problem.

## What does the law actually ask a report to contain?

In England the law asks for particulars, and that one word settles most of the argument.

Under [the Education (Pupil Information) (England) Regulations 2005](https://www.legislation.gov.uk/uksi/2005/1437/body/made), the head teacher, not the class teacher, prepares a report for every registered pupil before the end of the summer term and provides it to parents free of charge. Schedule 1 says what it carries. The first item is "brief particulars of achievements in all subjects and activities forming part of the school curriculum." The second is "comments on general progress."

Particulars. Not a description of the subject, not an account of what a child at that grade typically does, but particulars of what this one achieved. And the duty sits with the head teacher, which is the part leaders tend to forget. A report is not a teacher's private note that the school happens to post. It is the school's document, owed by the school to the parent, prepared in the head teacher's name. That is why the rule about how it gets written belongs to leadership rather than to individual taste.

A US district's report card sits inside state and local policy instead of this instrument, and the wording differs. The principle travels: a report is a statement a school makes about a named child to that child's family, and its worth rests on being about that child.

## Why is "the teacher checks it" the wrong line?

Checking is the wrong line because a reader cannot detect a plausible invention about a child they teach, and a parent cannot detect one at all.

This is where most school rules land, and it sounds like the responsible answer: let staff use the tool, require them to read the output. The trouble is what checking catches. A teacher scanning a draft will catch a wrong name, a wrong grade, a wrong subject. She will not reliably catch the sentence that is merely reasonable. Faced with "Ayesha has grown in confidence when presenting to the class," on report twenty-two of thirty, at nine in the evening, the honest answer is often that it sounds about right. Sounding about right is the standard the law declines to set.

Then there is the asymmetry that makes this a leadership question rather than a matter of professional standards. The parent is the only reader who cares about the particulars and the only one with no way to test them. A parent who reads that their quiet daughter has grown in confidence carries that sentence into the next parents' evening, into a conversation at home, sometimes into an option choice. They are entitled to assume a person saw it happen.

A sharper irony sits in the government's own paperwork. The Department for Education's [Generative AI: product safety standards](https://www.gov.uk/government/publications/generative-ai-product-safety-standards/generative-ai-product-safety-standards), published January 19, 2026, set out thirteen areas edtech suppliers are expected to meet. One asks a product to "detect cognitive offloading actions that indicate the learner is asking the system to do the work for them," with examples including pasting text into an answer box instead of writing your own response. Those standards are written for suppliers, and about children. Nobody has written the equivalent expectation for the adult pasting a grade into a box and accepting what comes back. If outsourcing the thinking is the habit we are teaching pupils to resist, the report is an awkward place for the staff to practice it.

## What is the Three Sources Test?

The Three Sources Test asks where each sentence in a report came from, because only two of the three possible answers may be sent.

> The Three Sources Test is the question I ask leadership teams to settle before AI goes anywhere near a pupil report. Every sentence a parent reads comes from one of three sources, and only two of them may go home. 1. The teacher's own observation of that child. 2. The school's own records: marks, assessment data, attendance, prior reports. 3. The model, which has never met the child and will supply a plausible particular to fill any gap the first two leave. Sentences from sources one and two may be drafted, shortened, smoothed or translated by AI. Sentences from source three may not be sent. The decision a leader has to make is therefore not whether staff may use AI on reports, but what a teacher must supply before the tool runs, so that source three has no gap to fill.

I offer it as a way of ordering the decision rather than a tested standard, and its use is diagnostic. Ask a teacher to read her own drafted paragraph and say, sentence by sentence, which source each came from. Where she can answer, the report is hers and the tool helped her write it. Where she hesitates, that sentence is this article's whole subject, and it is still in the document.

The test also says what to do about a gap, which is the useful part. If a teacher cannot name three particulars for a child, the model should not be invited to guess. A child nobody has looked at closely enough is a more valuable thing for a school to find than any report.

## Which parts of a report should AI write?

AI should write the parts of a report that were never about the individual child, and in most reports that is a larger share than leaders expect.

Open any set of reports and much of the text is structural: the sentence framing what the subject covered this term, the standard explanation of what a grade means, the target drawn from a shared bank, the closing line. None of that is a particular. It is the same for thirty children, it has to be written anyway, and it is where the evening actually goes. A model that turns shorthand into full sentences, trims a paragraph to fit the field, matches the house tone, or produces a version in a family's first language is doing work worth having.

Two honest limits. In the Bett survey, 51 percent of teachers said AI had eased their workload but only 35 percent said they were working fewer hours, and 55 percent reported the same hours as before. Workload relief and hours returned are different claims, and the second is the one a leadership team gets judged on. Separately, which pupil information may be typed into which tool is its own decision, and [whether staff may put student data into ChatGPT](https://blog.theaieducator.io/posts/can-staff-put-student-data-into-chatgpt) has to be settled first.

## What I Tell Leadership Teams

The rule most schools write is about the tool. The rule they need is about the input.

Across the leadership teams I work with, the mistake I see most often here is a single line in a policy that names an approved product and stops: staff may use the school's approved AI for report writing. That line permits everything this article warns about, because the risk never lived in the choice of product. It lives in what a teacher may hand over, and in whether anyone has said that a grade is not enough.

The second pattern is quieter and more common. A leadership team believes it has already settled this. [Teacher Tapp's survey on using AI to write reports](https://teachertapp.com/articles/can-you-use-ai-to-write-reports-plus-book-scrutiny-swimming-and-ceo-pay/), published June 10, 2025, found that 27 percent of head teachers thought their school allowed AI with specific rules, while only 11 percent of classroom teachers said those rules existed. More than half of the primary schools surveyed had no report-writing policy at all, a third allowed AI with no rules attached, and 5 percent had banned it. That gulf is the finding to carry into your next leadership meeting: a rule nobody downstairs can state is not a rule, it is a belief held upstairs.

Nor is the gap an English peculiarity. The RAND Corporation's survey panels, [published September 30, 2025](https://www.rand.org/pubs/research_reports/RRA4180-1.html), found 45 percent of US principals reporting school or district policies or guidance on AI use. Which leaves the majority of parents on both sides of the Atlantic with no way of knowing what their school permits, and no reason to ask until something reads oddly. If you are deciding what to tell families first, [how to talk to parents about AI](https://blog.theaieducator.io/posts/how-to-talk-to-parents-about-ai) sets out what belongs in that letter.

I write a daily podcast and a newsletter for school and trust leaders, and report writing is among the few AI questions that arrives from every direction at once: from teachers who have started, from head teachers who suspect it and have not asked, and from trust executives who want one answer for forty schools. It travels because nobody can see it happening. That is also why it needs a written rule rather than a shared understanding.

## Six lines to settle before this term's reports

Settle these six in one meeting, write them down, and send them to staff before anyone starts drafting.

1. **The floor.** The minimum a teacher supplies before using a tool: three specific observations per child per subject, or whatever number fits your reports. Name the number. "Enough detail" is not a floor.
2. **The permitted jobs.** What the tool may do with that material: expand shorthand into sentences, shorten to fit, match the house tone, translate. Write the list as jobs, not as a product name.
3. **The forbidden job.** Writing in a specific the teacher did not supply. One sentence, because this is the line and the rest is detail.
4. **What goes in.** Which pupil information may be typed into which tool, and what never may: names, assessment data, anything touching special educational needs or health. That is a data decision, and it should already exist.
5. **Who signs.** The named person who confirms the particulars are the school's own, by subject or by class. The statutory duty sits with the head teacher, so the chain has to end somewhere specific.
6. **The check.** One sampled set of reports per year group, read against the teacher's own notes rather than for typos. Agree who reads them and when, before reports go out and not after a complaint.

Read the answers together and you will see which of three positions your school is in. All six filled in is a decision. Lines one to three filled in and four to six blank is a rule with nobody behind it. Line three blank is where most schools sit without knowing it: the model is writing the particulars, and no adult has said that it may.

## When is a lighter rule enough?

A lighter rule is enough where the report was never carrying a particular.

The counterargument deserves a hearing, because a rule that treats every line of every report as a judgment call will be ignored by the second week. Three cases allow a lower floor. A secondary subject report consisting of a grade, an effort code and one sentence from a departmental bank is mostly source two, and can be drafted from the data with light review. A translation of a report already written and signed changes no particular, only the language. A tool built inside your management information system, able to read only your own records, has no gap to invent into, which is a real argument for the narrow tool over the general chatbot.

What a lighter rule never covers is the free-text paragraph about a named child. Parents read that part first, keep it, and quote it back to you, and it is the only part of a report that could not have been written by a stranger.

## What should leaders do this week?

Run the Three Sources Test on ten reports before writing any policy. Read each free-text paragraph and ask the teacher which source each sentence came from. Ten reports will tell you more about where your school stands than a term of drafting, and if the answers come back clean you can write a short rule with confidence instead of a long one out of anxiety.

Then settle the six lines and send them out. Where staff are already doing this without permission, that is a practice to bring inside the rules rather than a disciplinary matter, and [what to do about shadow AI](https://blog.theaieducator.io/posts/what-to-do-about-shadow-ai) covers how without driving it further underground. Where something has already gone home that should not have, [what leaders should do when staff get AI wrong](https://blog.theaieducator.io/posts/what-to-do-when-staff-get-ai-wrong) is the sequence. When the rule is agreed, put it in the policy rather than an email, because [what a school AI policy should contain](https://blog.theaieducator.io/posts/what-should-a-school-ai-policy-contain) is where staff will look for it next year.

One thing worth saying out loud in that meeting. If the floor feels high, that is the finding and not an objection. A school that cannot name three particulars per child has a problem its reports were quietly covering, and better sentences will not solve it. [Building capability rather than dependence](https://blog.theaieducator.io/posts/build-ai-capability-not-dependence) is the longer version of the argument. Reports are where it turns concrete.

## The next step

If your staff are already using AI on reports and you want one rule that holds across a school or a trust, this is the kind of work I support through [AI training for schools](https://theaieducator.io/ai-training-for-schools?utm_source=blog.theaieducator.io&utm_medium=referral&utm_campaign=should-ai-write-school-reports), where the decision and the practice get settled in the same room. For the wider picture, [why AI adoption is really a leadership problem](https://blog.theaieducator.io/posts/why-ai-adoption-is-really-a-leadership-problem) sets out why a question like this one lands on a leader's desk rather than a teacher's. If you would rather read than book anything, I write a [newsletter for school and trust leaders](https://theaieducator.io/?utm_source=blog.theaieducator.io&utm_medium=referral&utm_campaign=should-ai-write-school-reports#newsletter) on the decisions this kind of work turns on.

## Sources and further reading

- Schools Week, "AI use 'cuts teacher workload but not hours worked', study suggests," August 27, 2026, reporting a Bett survey of 1,033 teachers conducted by YouGov in June 2026. [schoolsweek.co.uk](https://schoolsweek.co.uk/ai-use-cuts-teacher-workload-but-not-hours-worked-study-suggests/)
- The Education (Pupil Information) (England) Regulations 2005, SI 2005/1437, regulation 6 and Schedule 1. [legislation.gov.uk](https://www.legislation.gov.uk/uksi/2005/1437/body/made)
- Department for Education, "Generative AI: product safety standards," GOV.UK, January 19, 2026. [gov.uk](https://www.gov.uk/government/publications/generative-ai-product-safety-standards/generative-ai-product-safety-standards)
- Teacher Tapp, "Can you use AI to write reports? Plus Book Scrutiny, Swimming, and CEO Pay," June 10, 2025. [teachertapp.com](https://teachertapp.com/articles/can-you-use-ai-to-write-reports-plus-book-scrutiny-swimming-and-ceo-pay/)
- Karen Wespieser, "From novelty to normal: how teachers are using AI in 2026," Chartered College of Teaching, May 18, 2026, reporting Teacher Tapp survey data from November and December 2025. [my.chartered.college](https://my.chartered.college/impact_article/from-novelty-to-normal-how-teachers-are-using-ai-in-2026/)
- Christopher J. Doss and others, "AI Use in Schools Is Quickly Increasing but Guidance Lags Behind: Findings from the RAND Survey Panels," RAND Corporation, September 30, 2025. [rand.org](https://www.rand.org/pubs/research_reports/RRA4180-1.html)

*Dan Fitzpatrick is the founder of The AI Educator, a Forbes contributor and the author of bestselling books on AI in education. [Read more about Dan](https://theaieducator.io/about?utm_source=blog.theaieducator.io&utm_medium=referral&utm_campaign=should-ai-write-school-reports).*

## Frequently asked questions

### Can teachers use AI to write school reports?

Yes, for the prose, provided the teacher supplies the particulars first. AI can expand a teacher's shorthand into sentences, trim a paragraph to fit, match the house tone or translate it. What it must not do is invent a specific about a named child that the teacher did not supply.

### Is it against the rules to use AI for pupil reports?

No national rule in England or the United States bans it. The duty is about content, not method: under the Education (Pupil Information) (England) Regulations 2005 the head teacher's report must carry brief particulars of a pupil's achievements. A school sets its own rule on how those particulars get written up.

### Does the Department for Education allow AI for report writing?

The DfE has not prohibited it. Its Generative AI: product safety standards, published January 19, 2026, are written for edtech suppliers rather than schools, and ask products to detect when a learner is offloading the thinking. No equivalent expectation has been written for staff drafting reports.

### Should schools tell parents if AI helped write a report?

Decide it deliberately rather than by default. If your rule is that teachers supply every particular and AI only shapes the prose, that is a straightforward thing to tell families, and saying it first is easier than answering it later when a parent asks whether a person wrote what came home.

### Can teachers put pupil names into ChatGPT to write reports?

That is a data decision and it should be settled before the report question. Which pupil information may be typed into which tool, and what must never be, including anything touching special educational needs or health, belongs in the school's policy on staff use of AI tools.

### What should a school's rule on AI and reports actually say?

Six things: the minimum a teacher supplies before using a tool, the jobs the tool may do, the one job it may not, which pupil information may be entered, who signs off that the particulars are the school's own, and who samples a set of reports against the teacher's notes.

### How do you stop AI reports sounding generic?

Generic prose is a symptom rather than the problem. A report reads generically when the model was given nothing specific to work with. Require three concrete observations per child per subject before drafting, and the writing fixes itself, because the particulars were there all along.

---
Source: [Should Teachers Use AI to Write School Reports? Where to Draw the Line](https://blog.theaieducator.io/posts/should-ai-write-school-reports)
Publisher: [The AI Educator](https://theaieducator.io)