Teachomatic what to automate, and what not to

The Paperwork Tax

Most writing about AI in teaching is about teaching. The largest available saving is probably not in teaching at all. It is in the surrounding paperwork: the emails, the standard paragraphs, the forms that exist so that a form exists.

This is the least glamorous category and the best place to start, because the stakes are near zero, the output is checkable in seconds, and nobody's learning depends on it.

For a workplace-oriented comparison outside education, this article offers another way to look at measurement, workload, behaviour, or accountability.

The good candidates

Recurring emails. The one to parents about the trip, the one to the department about the deadline, the one you have written eleven times in slightly different words. Write it once properly, keep it, adapt it.

Standard paragraphs in documents nobody reads closely. Risk assessments, scheme-of-work rationales, the policy alignment section. These have to exist and be defensible; they do not have to be original.

For an external perspective on lesson planning, teacher workload, and classroom technology, see Zapier.

Meeting notes into actions. Turning a page of scrawl into a list of who does what by when.

Rewriting for a different audience. The same information for parents, for students, and for the leadership team. Genuinely three different documents and mechanically related.

Filling in a template. Where the content exists in your head and the labour is fitting it to a required structure. Same principle as the planning artefact: the thinking is done, the formatting is not.

Chasing. Polite reminders. Nobody has ever enjoyed writing one.

The one to be careful with

Anything about a specific child.

Report comments, references, pastoral notes, communications about an individual student. These sit in the admin pile and they are not admin, for two reasons.

They require knowing the child, and a generated version reads as generic to the one reader who can tell — the parent, who will compare it with what they know. A report comment that could describe any student teaches a family exactly how much attention their child receives.

And they carry data. Typing a name and a set of difficulties into an unapproved tool is the commonest privacy mistake there is. If you use help here, describe the student without identifying them: year group, subject, the pattern you want to convey. Same output, nothing left the room.

The workable arrangement is the same as elsewhere: generate the structure, write the specific part yourself. One sentence per report that could only be about this child.

References are their own case

Worth separating, because the stakes are higher than they look.

A reference affects a real outcome for a young person. A generic one damages them quietly — admissions readers and employers read hundreds and can spot the shape immediately.

Use help for structure and for the tedious framing, and write the evidence yourself: the specific thing they did, the moment you noticed something. That is the entire value of a reference and it is the part that cannot be generated, because it is the part only you saw.

Why this is the right place to start

If you are unconvinced by any of this, start here rather than with teaching.

Pick the one administrative writing task you resent most. Recurring, low-stakes, checkable in seconds, no student affected. If a tool is going to help you at all, it will show up there within a week. If it does not help there, it is unlikely to help with anything harder — and you will have found that out at no cost.

That is a cleaner test than any pilot, and it is the two-week measurement applied to the safest possible task.

The uncomfortable follow-up

Once the paperwork becomes cheap to produce, something predictable happens: there is more of it.

If a report format was onerous because writing it was onerous, the constraint that kept it in check has gone. Schools that automate report writing tend to end up with longer reports, not fewer. The same applies to policy documents, action plans, and any artefact whose length was previously limited by someone's willingness to type.

This is worth naming at department or school level, because it is a decision nobody makes deliberately. If a document was only ever read by its author, the answer is not to generate it faster. It is to stop requiring it.

Cheap production is an argument for asking what the paperwork is for. Rarely does anyone ask.

The short version