Most of what you believe about good feedback was shaped by a constraint you never chose. Feedback was scarce. There was one of you and thirty of them, and your red pen moved at the speed of a human hand at the end of a long day. So the rules of “good feedback” that you absorbed during training — be selective, prioritise, don’t over-comment, focus on the two things that will move the grade — were really rules for rationing a scarce resource well.

That constraint is dissolving. AI-guided feedback can now return an examiner-style response to an IGCSE answer minutes after a student writes it, to every student, at no marginal cost to your evening. And when a constraint that fundamental disappears, the definition of “good” built on top of it has to be re-examined. This is a reflective piece, not a sales pitch: I want to think through what feedback in IGCSE & A-Level classrooms actually becomes when it stops being scarce — and, just as honestly, what it shouldn’t become.

The old definition was a workaround

It’s worth being precise about what we used to mean by good feedback, because most of it was a workaround for scarcity rather than a principle.

“Be selective” was good advice because you couldn’t comment on everything, so you’d better choose well. “Return work promptly” was an aspiration constantly defeated by the maths — thirty scripts, six classes, a fortnight before you handed them back, by which point the lesson had moved on and the student had emotionally filed the work away. “Make comments actionable” was sound, but actionability decays with time: a precise note about a paragraph the student wrote two weeks ago lands on someone who no longer remembers writing it.

None of those were wrong. But notice that all three were responses to the same underlying scarcity. Take the scarcity away and the advice doesn’t simply stay true — some of it inverts.

What the research actually said — and why AI changes the economics

The feedback research has been remarkably consistent for decades, and it’s worth restating because AI doesn’t overturn it — it makes parts of it suddenly affordable. Good feedback, the evidence says, is timely, specific, and actionable. It works best as a loop rather than a verdict: the student does something with it and you respond to what they did. Hattie and Timperley’s framing — where am I going, how am I doing, where to next — assumes a conversation, not a one-way delivery of marks.

Here’s the quiet tragedy of the pre-AI classroom: we knew all of this, and we couldn’t do most of it. Timely was impossible at scale. The feedback loop — comment, revise, re-comment — was a luxury reserved for coursework and the occasional redraft, because closing the loop meant marking the same work twice. So the research described a gold standard that the economics of one teacher and thirty students quietly forbade.

AI-guided feedback changes those economics, not the principles. Timeliness becomes free. Specificity at volume becomes possible — a system anchored to the mark scheme can tell every student precisely which marking points they hit and missed, on every answer, without fatigue. And the loop becomes affordable for the first time: a student can answer, get examiner-style feedback, revise, and get feedback again before they’ve left the topic. The things the research always told us were valuable but rationed are exactly the things automation makes abundant.

When feedback is abundant, scarcity moves somewhere else

This is the part worth sitting with. When something becomes abundant, value doesn’t disappear — it relocates to whatever is still scarce. Cheap printing didn’t make writing worthless; it made editing and curation the valuable part. Abundant feedback works the same way.

When every student can get instant, specific, mark-scheme-accurate notes on what they did, the scarce thing is no longer the correction. The machine can tell a student that their “evaluate” answer asserted rather than weighed, that they never reached a judgement, that two marking points went unaddressed. What it cannot reliably do is judge what this particular student most needs to hear right now, model the move from a Band 3 argument to a Band 5 one in a way that clicks for them, or notice that a usually-fluent writer has gone strangely flat and ask why.

So the teacher’s comment shifts. It moves away from marking up errors — the machine now does that faster and more consistently than a tired human ever could — and toward three things the machine can’t do: directing attention, modelling thinking, and having the conversation.

What the teacher’s comment becomes

From marking up errors to directing attention

A student facing a wall of accurate, instant feedback has a new problem: not too little information, but too much. Fifteen flagged points across an essay is noise. The genuinely human, high-value act is selection — of these fifteen things the system caught, this one matters most, fix it first, the rest will follow. That’s not marking; that’s coaching. The scarcity has moved from generating the feedback to prioritising it, and prioritising for a specific learner is a judgement call AI is poorly placed to make.

From correcting to modelling thinking

The machine is excellent at identifying that a marking point is missing. It is far weaker at the thing that actually teaches: showing the thinking that would have produced it. When you write “you’ve described the trend but not explained the mechanism,” and then think aloud — here’s how I’d have reached for the mechanism, here’s the question I’d ask myself — you’re modelling a cognitive move, not flagging an absence. That demonstration of expert reasoning, made visible, is the scarce resource now. It’s what distinguishes a comment that fixes one answer from one that changes how a student approaches the next ten.

From verdict to conversation

The most underrated effect of abundant feedback is that it frees you to be the second voice rather than the only one. Let the system handle the first pass — the “here is what an examiner would credit and dock.” That clears your attention for the conversation the machine can’t have: why did you make this choice? what were you trying to argue here? you nearly had a brilliant point — what stopped you? Feedback as dialogue was always the research’s gold standard and always the first thing sacrificed to time. It’s the part of the job most worth protecting, and it’s now the part AI gives you room for.

The honest counterweight

I don’t want to oversell this, because there are real risks and pretending otherwise would be the kind of hype this blog tries to avoid.

Abundant feedback can become abundant noise. A student drowning in instant corrections, with no one helping them see which matter, may end up learning less than one who got a single sharp comment a fortnight late. Volume is not value. There’s also a quieter risk: that students learn to optimise for the machine’s checklist — chasing marking points — rather than learning to think, which is exactly the failure mode I worry about most and have written about separately in using AI feedback without dumbing down teaching.

And there’s a risk to us. If the machine writes all the feedback, a teacher can drift out of the loop and lose the diagnostic feel that comes from reading student work closely. The calibration that hand-marking gave us — that sense of where a cohort really is — is worth preserving even when you’re not the one doing the first pass. The goal isn’t to stop reading student writing; it’s to stop spending your reading on the parts that never needed a human.

So what does “good feedback” mean now?

If I had to rewrite the definition for the AI era, it would be something like this. Good feedback is no longer measured by how thoroughly you marked up the errors. The thoroughness is assumed — abundant, instant, consistent, handled. Good feedback is now measured by whether the whole loop worked: did the student get something specific while it still mattered, did a human help them see what to do first, did a conversation happen that the machine couldn’t have, and did the student come back and try again.

The teacher’s comment doesn’t get less important in this picture. It gets more important, because it’s no longer competing with the clerical work for your attention. For a closer look at what the automated layer can and can’t do well, what examiner-style AI feedback looks like and what AI marking gets right are the honest companions to this argument; for the bigger anxiety underneath it all, will AI replace teacher marking tackles it head-on.

If you want to see how this division of labour feels in practice, Tutopiya’s free platform for teachers gives examiner-style AI feedback against the actual Cambridge and Edexcel mark schemes — instantly, with a review-and-override step so the first pass is the machine’s and the judgement stays yours. It’s the cleanest way I know to test whether moving your attention up the chain actually makes your feedback better. (It’s free to start with one class.)

FAQ

What does “AI-guided feedback” actually change about good feedback? It changes the economics, not the principles. Good feedback was always meant to be timely, specific, actionable, and conversational — but scarcity made most of that impossible at scale. AI makes the timely-and-specific layer abundant, which moves the teacher’s value toward prioritising, modelling thinking, and having the dialogue the machine can’t.

Does this mean AI writes the feedback and I just approve it? No — and that’s the version to avoid. The useful model is that AI handles the first pass (what an examiner would credit and dock), which frees your attention for the second, harder layer: helping the student see what matters most and why. The human comment becomes more important, not less.

Won’t abundant feedback just overwhelm students? It can. Abundance without prioritisation is noise. That’s precisely why the teacher’s role shifts toward selection — telling a student which of the flagged issues to fix first. Volume is not the same as value, and managing that is now part of the job.

Is there a risk students just chase the machine’s mark scheme instead of thinking? Yes, and it’s the risk worth watching most closely. The defence is keeping a human in the loop to reward genuine reasoning — including valid arguments the mark scheme didn’t anticipate — rather than letting checklist-chasing become the goal.

Does AI in education make the teacher’s feedback less valuable? The opposite, used well. It removes the clerical marking that was never worth your judgement and leaves you the parts that genuinely need a teacher — directing attention, modelling expert thinking, and the conversation. The scarce, valuable thing was never the correction; it was the human read on what to do with it.

The bottom line

The constraints that defined good feedback — scarce, slow, hand-written — are dissolving, and that’s not a threat to the teacher’s comment. It’s a clarification of what it was always for. Let the machine make feedback abundant, and spend your now-undivided attention on the part that was always the point: helping a specific student see what to do next, and showing them how an expert thinks.

See how the first-pass-by-AI model works, free with one class →

Ready to Excel in Your Studies?

Get personalised help from Tutopiya's expert tutors. Whether it's IGCSE, IB, A-Levels, or any other curriculum — we match you with the perfect tutor and your first session is free.

Book Your Free Trial