Teachomatic what to automate, and what not to

When Work Changes Suddenly

A student who has been handing in adequate work submits something noticeably better. Different vocabulary, tighter structure, an argument that holds.

This is the most common trigger for suspicion, more common than any detector score, and it is a weak signal — because improvement looks exactly like the thing you are worried about. That is not a coincidence. It is definitional: you are reacting to a gap between the work and your expectation, and every route to closing that gap produces the same appearance.

A workplace-oriented comparison is Monitask’s discussion of how Microsoft Teams tracks activity, which offers another context for thinking about workload, behaviour, or accountability.

The explanations, roughly by frequency

Before assuming the obvious one, the field is wider than it feels at 11pm with a stack of marking.

They tried. The most common and the most overlooked. A student who has been coasting engaged with a topic that interested them. If you treat this as suspicious, you punish exactly the behaviour you spent the term asking for.

For an external perspective on academic integrity, technology, and student rights, see American Psychological Association.

They got help. A parent, a sibling, a tutor, a friend, a homework club. This has always happened, has never been evenly distributed, and mostly is not called cheating.

They finally understood something. Learning is not linear. Students plateau and then jump, particularly when a foundational misunderstanding clears.

They had time. Their usual work is produced in twenty minutes on a bus. This one was done on a free weekend.

They copied a model. They found a good example and followed its structure closely. Somewhere between learning and plagiarism, extremely common, and usually a teaching moment.

They used a tool and did not declare it. Real, and one of six, not the default.

They used a tool and were allowed to. Check what you actually permitted for this task before anything else. If your own rules were unclear, that is on the rules.

What is actually diagnostic

Not the quality. Quality is the weak signal. What carries information:

Discontinuity within the piece. One paragraph in a different register, or a section that is suddenly assured surrounded by their normal writing. Human improvement is usually uniform; patched work is not.

Content you did not teach and they could not have. A framework, a term, a reference from outside the course. Not proof — students read — but worth asking about.

Answering a question you did not set. Slightly adjacent to your actual task. This happens when a prompt was paraphrased.

Fabricated specifics. A citation to something that does not exist, a quotation not in the text, a statistic with no source. This is the characteristic failure mode, and it is checkable in seconds.

Absence of their errors. You know their habitual mistakes. Their sudden disappearance, alongside a new set of unfamiliar ones, is more informative than the quality.

Note that all five are specific and checkable. "It doesn't sound like them" is not on the list, because that is where the errors concentrate on students who write plainly or in a second language.

What to do

Nothing dramatic, and nothing silent.

The wrong move is to say nothing and mark it down, or to say nothing and file a suspicion for later. Both leave the student unable to respond to something that is affecting them.

The right move is a two-minute conversation about the work. This is a real step up from your last piece — talk me through it. What made the difference?

Notice what that question does. If they tried, it is praise and they will tell you what changed. If they got help, they will usually say so, because you have not framed it as an accusation. If they used a tool, the answer is often an admission, because you have made it easy. And if they wrote it and cannot discuss it — which is rare — you have learned something.

Ask it of good work generally, not only when you are suspicious. If the only students you ask are the ones you doubt, the question becomes an accusation with a friendly face, and students work that out in about a week.

If you find undeclared use

Go to the procedure, and hold onto one thing: a sudden improvement plus undeclared assistance is usually a student who was behind, panicked, and did not know where the line was.

The proportionate response is the line, clearly, with a consequence that matches. Not a case.

The thing this costs if you get it wrong

A student who improves and is met with suspicion learns that effort produces scrutiny. That lesson is durable and it is precisely backwards.

This risk is not evenly distributed either. It falls on students whose previous work set a low expectation — which is to say, the students for whom improvement mattered most.

So: ask about the work, ask everyone, and start from the assumption that it went well. You will occasionally miss something. The other error is worse and it is the one you will not find out about.

The short version