Teachomatic what to automate, and what not to

Feedback a Student Will Read

The problem with feedback was never that there was not enough of it. Teachers have been writing comments that students glance at and discard for as long as there have been comments.

Automation makes this worse in a specific way: it removes the constraint that used to force selection. When writing feedback cost you four minutes a student, you had to decide what mattered most. When it costs thirty seconds, you get comprehensive coverage of everything — and comprehensive coverage is the format students most reliably ignore.

For a workplace comparison of the underlying concept, Monitask’s discussion of task switching cost shows how the same behaviour is framed outside education.

So the question is not how to produce more feedback faster. It is what makes a student act on any of it.

What students act on

One thing. Not five. A student who reads five points addresses none of them, because there is no obvious place to start and the implicit message is that the work was broadly wrong.

For an external perspective on teaching, assessment, and feedback, see Cult of Pedagogy.

Something they can do. "Develop your analysis" is a description of the gap, not an instruction. "In your second paragraph you say the writer uses short sentences — add one sentence saying what effect that has" is a thing a person can do on Tuesday.

Something specific to their work. The moment feedback could have been written without reading their piece, it stops carrying any signal, and a student can tell in under a second.

Something that acknowledges the attempt. Not praise sandwiches — recognition that you saw what they were trying to do, especially when it did not work. A student who tried something ambitious and failed has done better work than one who played safe, and feedback that treats both as errors teaches the wrong lesson.

A reason to look, next time. Feedback is a relationship over a term. If a student has learned that your comments are worth reading, they read them.

What automation does well

Mechanics and structure. Accurate, fast, and genuinely tedious to write by hand.

Criteria mapping. Where the work sits against the rubric, in complete sentences. Useful, and the part nobody enjoys writing — though rubric words like "insightful" are placeholders for judgement rather than criteria.

A first draft to cut down. This is the real use. Generate, then delete four fifths of it.

What it cannot do

Everything that depends on knowing this student — and that is not a soft claim, it is an information problem. The tool does not have the data.

It does not know you told them this three weeks ago. It does not know this is the first time they have attempted an argument at all, and that the attempt is the achievement. It does not know they are capable of much more and are coasting. It does not know something is going on at home.

This is why generated feedback reads as accurate and lands as nothing. It is correct about the work and blind to the person, and students experience that combination as being processed rather than read.

The workable method

Generate the full analysis and do not send it. Read it, mostly to catch things you missed.

Choose one thing. The one that will move this student furthest, which is a judgement about them, not about the work. Frequently it is not the biggest error.

Write that in your own words, referring to a specific place in their piece. Two sentences.

Add one sentence that only you could write. This is the first time you've disagreed with a source and I want more of it. You've fixed the thing we talked about last time. This is well below what you did in October — is everything alright?

That last sentence is the whole game. It takes fifteen seconds and it is the only part the student is certain to read.

The efficiency trap

Here is the thing to watch, and it is the same pattern the research keeps finding: saved time turns into more feedback rather than less work.

More feedback is not the improvement it sounds like. Thirty students receiving four paragraphs each is worse than thirty students receiving three sentences each that they act on, and it costs you more.

So decide what the saved time is for before you have it. Deeper feedback for the students who need it most is a good answer. More words for everybody is not.

Feedback they cannot act on is worse than none

Worth stating plainly, because it is the failure mode automation makes cheap.

A student who receives detailed, accurate, generic feedback learns something specific: that producing work leads to a wall of text nobody wrote for them. That is a lesson about how much their effort is worth, and it is more durable than anything in the comments.

Three sentences that are obviously about their work will do more than a page of correct observations. If you only take one thing from this page, take that — and note that the diagnosis you do while reading their work is where those three sentences come from.

The short version