How to write reports and family communications with AI

TLDR: Term reports and family emails consume writing hours that are not teaching hours. MagicSchool generates reports and communications from your notes inside its free plan, Claude helps find the tone for difficult emails, and Gamma formats the newsletters. The raw material is still your concrete observations: a report generated without real notes is a template families detect.

Report card number nine

You are on the ninth of twenty-eight and you have just reread it. It says the student “shows a positive attitude and participates with interest, though consistency in daily work could improve”. That is word for word what the third one said. The worse part is not the repetition. It is that both times it was true and both times it is useless, because a family reading that has no idea what to do on Monday.

That is the actual work this guide is about, and it is not writing. When a report comes out generic it is almost never because you cannot write. It is that by the ninth one you have run out of concrete facts and are running on impressions, and a carefully worded impression is still an impression. AI speeds up the writing, which is the cheap part. What decides whether the report is any use is what you have written down about that person over the previous ten weeks.

AI supplies the writing and you supply the observation. A report without concrete data about the student is institutional filler, however it was written.

1. Prepare concrete notes on each student

The report families appreciate contains things only you know: what their child does well with examples, exactly where they get stuck, what change showed this term and what is worth working on at home. None of that can be invented by a tool, and when it is invented it shows.

The habit that feeds this flow is the brief weekly note: two lines per student who stood out that week, in whatever tool. At term’s end, those accumulated notes are the raw material for reports that sound like the real person, not the job title.

A note is worth keeping if it has a date and a fact. “On the 14th she explained decimal place value to her partner better than I did” is worth keeping. “Good attitude” is not, because nine weeks later you will not be able to reconstruct what you saw. And the spread matters as much as the content: if some students have six notes and others have none, the class has quietly become the ones who shine and the ones who disrupt, with everybody else invisible. Those are exactly the reports that come out empty.

If you reach the term without notes, spend one session writing three concrete sentences per student before generating anything. It is the entire difference in the result.

2. Generate the individual reports in batches

MagicSchool includes tools built for teacher communication: give it your notes on each student and the report type, and it returns the text with professional structure and appropriate tone. Inside its free plan, with the paid one from $8.33 a month.

There is a fast test for the whole batch, and it belongs before anything gets sent. Put all twenty-eight reports on one screen and look for the most repeated sentence. If it turns up in more than a third of them, you did not write twenty-eight reports, you wrote one with twenty-eight names on it. The cause is always in step 1, and so is the fix.

The report describes what was observed. The grade has its own criterion and its own defence, and that gets prepared in exams and quizzes rather than here. A report that spends its length justifying a grade ends up as a badly written gradebook entry.

3. Handle the difficult emails with extra care

The email about a behavior problem, the reply to an angry family or the news of a poor trajectory are the pieces where AI helps most and demands most review. Claude (from $17 a month, with a permanent free tier) is the assistant for this job: ask it for the direct version and the diplomatic version of the same email and settle on the middle point. An email to an angry family is won or lost entirely on nuance, and that is where its writing beats anything else in the catalog.

The structure that works in the difficult email has three stretches and the order matters:

AI builds that scaffolding well. Choosing which facts to tell and what to propose is the professional judgment that does not get delegated.

There is a use for AI in this step that beats the drafting, and hardly anybody tries it: ask it to read the email the way the family will. Paste the draft and ask which sentence could land as an accusation, which part reads as though you have already decided something, and what questions the recipient will send back. It usually finds two, and they are usually the two you were about to send.

And the safety rule: the difficult email gets reread in full before sending, word by word. One sentence with the wrong nuance in a delicate conversation costs months of relationship with that family.

4. Build the newsletters and group communications

General communications (the field trip notice, the meeting summary, the start-of-year guide) demand less tact and more clarity. Gamma turns them into documents with readable formatting and visual hierarchy, which in a communication full of dates, times and steps matters more than it seems. Free credits to start, paid from $10. A newsletter has exactly one quality test: the key information (date, time, what to bring, what to sign) is findable in ten seconds.

The four exceptions that always show up

The flow works for most of the class. These four cases fall outside it, and each has its own exit:

The student you have nothing written about. They do not disrupt, they do not stand out, they hand things in. You reach their report and discover ten weeks with zero notes. Do not generate it: that is precisely the report a family reads as “nobody is looking at my kid”, and they would be right. The exit is short and uncomfortable: give them two sessions of deliberate observation before you write anything, and log the gap for next term.

The family that replies to argue. The report goes out and comes back as a long message rebutting it point by point. The temptation is to answer in writing with more evidence, and with AI open that takes four minutes. Do not. A disagreement about substance does not close over email, it closes in a meeting, and the reply’s only job is to arrange one without adding a single new fact.

The case that touches safeguarding or a diagnosis. Suspicion about a home situation, signs of a condition, anything that sounds like a clinical opinion. That leaves this guide’s flow completely. It does not get generated, it does not get pasted into any tool, and you do not communicate it on your own initiative: it goes through your school’s safeguarding or student support route, with its own formats and its own named people.

The student with more than one teacher. In secondary that is the default, and it produces the quiet failure: two teachers generate reports on the same student from different notes and different standards, and the family receives two versions that do not line up. It gets fixed before generating, by comparing the facts, not afterwards by editing the texts.

What to measure to know it is working

Three numbers you can count yourself at the end of term. None needs a tool and none is a percentage saved:

How many reports contain at least one dated fact. Open them and count. That is the number that separates this system from the template you already had, and the first time it usually stings.

How many families reply asking what you meant. A report that lands generates few “could you clarify this?” messages. It is the cleanest signal the text got through, and it compares term against term.

Minutes per report, timed in two batches. Time five reports early in the term and five at the end. If the time does not drop but the quality rises, the system is working too, the saving just moved from writing into observing.

Who signs what

The boundary here is short and takes no qualifiers: every report is read and signed by the person who teaches that student, one at a time, with their own name on it. No batch generation goes out without that individual read, however happily the tool exports all of them at once. The day a batch goes out unread, whatever error it contains still carries your name.

Same with the three things AI cannot decide: which facts get told, what gets left out on purpose, and what gets proposed. That is professional judgment, and it is the entire reason the report exists.

Underneath all of it sit your school’s data protection rules, which outrank this guide and any tool. If your school has nothing written down about what may be pasted into a generator, that conversation comes before building anything.

This flow connects with the rest of the teaching work: the weekly notes from step 1 come almost free if you use the continuous assessment from the interactive lessons guide, the full materials cycle is in AI lesson planning, and if what you need is content adapted to one specific student, that lives in differentiated materials. The whole sector lives in AI for education and training.

Frequently asked questions

Will families notice the report was made with AI?

They notice an empty report, however it was written. With concrete observations about the student inside, the generated text sounds like a teacher who knows their group, because the content is exactly that.

When is AI the wrong tool for this?

For the delicate case that is already tense, and for any communication about safeguarding or health. Not because the text would be poor, it would be good, but because in those conversations the family is reading for whether a person is on the other end. A flawless message at the wrong moment reads as distance, and a three-minute phone call beats the best email of the term.

Can I use these tools with minors’ data?

With care: initials or first names when generating, no sensitive data, and your school’s data protection rules first, which sit above this guide and any tool.

What if my school does not allow AI with student information?

Then the flow runs on anonymized notes and the names go in afterwards, by hand, in the final document. You lose some of the convenience and none of the method, because the value comes from your observations and you write those either way. Ask before assuming which of the two things the policy actually forbids: it is usually the identifiable data, not the tool.

Does it work for whole-school communications?

The mechanics are the same at larger volume: newsletter templates, reports with a common structure and a defined institutional tone. There it pays for leadership to fix the style guide before every teacher generates on their own.