Guides / ChatGPT and report comments

Can ChatGPT write your report comments? An honest audit

Plenty of teachers are already doing it. It sort of works, it fails in predictable ways, and there is one line you must not cross. Here is the full picture, including the workflow that makes it actually useful.

Every staffroom now has someone quietly running their report comments through ChatGPT or Claude. The question is not really whether you can (you can), it is whether the output is good enough to send home, how much work it actually saves once you do it properly, and whether you are allowed to do it at all. We have used these chatbots on our own classes, so this is an audit from practice, not a hit piece: what they are genuinely good at, the workflow that makes them work, and where they fall down.

What a raw chatbot is genuinely good at

The workflow that makes it work

Here is the thing the "just use ChatGPT" advice skips: a chatbot knows nothing about your students. Every true, specific sentence in the output has to come from evidence you typed in. So the teachers doing this well have all converged on the same workflow, which is essentially manual data assembly:

  1. Go through the markbook and write an evidence note per student: strongest topic, weakest topic, anything notable. Some teachers literally do a post-it note per student, or a photo of a handwritten grid.
  2. Write one careful mega-prompt: the comment structure you want, length limits, tone, banned phrases, your school's conventions.
  3. Feed it the class in batches, then edit every single comment against your own knowledge of the student.

This works. The comments come out specific, because you supplied the specifics. But notice what happened to the time saving: you are still doing the evidence gathering (the slowest part of report writing), you have added a prompt-engineering step, and you must still edit everything. What the chatbot removed is the typing, which was never the expensive part. The honest description is: better comments for similar total effort, not the same comments for less.

Where it falls down

The minimum safety rules if you use a chatbot anyway: initials or first names only; no school name, no year-and-full-name combinations, no health or behavioural detail; turn off chat history/training where the setting exists; edit every comment against your own knowledge; and never send a line you could not defend to the family with your markbook open.

The honest verdict

A raw chatbot is a very good phrasing engine with no evidence and no memory of your standard. The workflow that fixes this (assemble evidence per student, prompt carefully, edit everything, keep names out of it) genuinely produces good comments, and if reports come around twice a year and you enjoy tinkering, it may be all you need. What it does not do is save the time you wanted saved, because the expensive parts of report writing were always the evidence gathering and the judgment, and those stayed on your desk.

We built the post-it workflow so you don't have to

Markpilot is that exact workflow, productised: your marked tests or rubric judgments become the per-student evidence automatically, every comment is drafted from the actual marks (with nothing invented to fill gaps), tone and standard hold steady across the whole class, first names are all it ever needs, and you edit and approve every line. The phrasing engine, with the evidence layer attached.

Start free 15 credits a month free · no card required · built by practising teachers