One Essay a Week vs Ten a Night: Why Feedback Speed Changes Everything

A law student writes a practice essay on contract law. They hand it in on Monday. They get it back, with comments, the following Monday.

By then they have forgotten which parts they were unsure about, why they structured it that way, and what they were trying to do in the third paragraph. The comments are accurate and almost useless, because the thinking they refer to is gone.

That week-long gap is the single biggest constraint on how fast anyone improves at written exams. Not talent, not hours, not the quality of the notes. The gap.

The loop, not the grade

Improving at anything performed — writing, playing, diagnosing, arguing — follows the same loop: attempt, find out what went wrong, attempt again with that in mind. How fast you improve depends on how many times you close that loop and how tight each turn is.

Most exam preparation never closes it at all. You read the material, you feel ready, you sit the exam, and the exam is the first time anyone checks. That is one turn of the loop, at the worst possible moment.

Practice essays close it once a week at best, and only if someone is willing to mark them. A professor with 120 students cannot mark ten essays per student. That is not a failure of teaching, it is arithmetic.

So the constraint was never the feedback's quality. It was its availability.

What changes when the gap collapses

Take the same law student. They write a practice essay and get it marked in under a minute: which principles they applied correctly, which they missed, which they applied wrongly. They read it while the essay is still in their head, see the gap, and immediately write another.

In one evening: five to ten essays, each one informed by the last. Under the old model that was two months of Mondays.

Three things change, and only one of them is speed:

Volume becomes possible. Ten attempts teach more than one attempt with better comments. This is the largest effect by far, and it is purely mechanical.

The correction lands on live thinking. You still remember why you wrote what you wrote, so the feedback attaches to a decision rather than to a piece of paper.

Failure gets cheap. When an attempt costs an hour of someone else's time, you write safe. When it costs a minute, you try the structure you are unsure about — and trying things you are unsure about is where the learning is.

The same shape applies outside law: medical students practising clinical reasoning, business students on case studies, engineers on design problems, anyone sitting an exam with written answers.

What the feedback has to contain

Speed alone does nothing. A score delivered instantly is still just a score, and a number tells you where you stand without telling you what to do.

Useful feedback does three things, in ascending order of value:

1. Says what was wrong. The minimum. Better than a mark alone, which most practice never gets.
2. Says why. Corrective feedback — explaining the error — consistently outperforms just revealing the right answer.
3. Connects it to the underlying idea. The correction that generalises: not "you missed the consideration requirement" but "you are treating formation requirements as a checklist rather than testing whether each one is satisfied on these facts".

If your feedback is stuck at level one, you will fix individual answers and keep making the same class of mistake. In ExamFlow, written answers are graded against your own material rather than against a generic rubric, which is what makes the third kind possible — the model can point at the part of your syllabus you skipped.

Where this breaks

Honest limits, because this is a technique with real failure modes.

The score is orientative, not official. Treat an AI mark as a rough position, never as a prediction of your grade. It is directionally useful and precisely wrong.

Subjective and creative work grades badly. For anything where the merit is in originality or voice, the feedback gets generic. It handles "is this argument complete" far better than "is this argument good".

It cannot catch what is not in your material. Grading against your own documents is a strength for relevance and a weakness for gaps: if your source material is missing something, neither you nor the model will notice it is missing. Past papers are the check for that.

Fast feedback can become passive. The failure mode of a tight loop is skimming the correction and immediately attempting again without thinking. The minute you spend deciding what you will do differently is the minute that does the work. Without it you are just generating volume.

How to actually use it

A routine that works, in ascending difficulty:

• Write cold. No notes open. If you consult the material while writing, you are testing your reading, not your recall.
• Read the feedback before rewriting, and name the error out loud in one sentence. If you cannot name it, you have not understood it.
• Rewrite only the part that failed, not the whole essay. Rewriting what already worked feels productive and teaches nothing.
• Keep a list of repeated errors. After ten attempts the list is short and it is the real syllabus of your weaknesses. This is what to take to your professor — a specific pattern beats "can you look at my essay".
• Every fifth attempt, use a past paper question instead of a generated one, to catch blind spots your own material cannot show you.

What to do when you disagree with the feedback

This comes up constantly and almost nobody addresses it. The grader says your answer missed something, and you think it did not. Or it marks a point as wrong that your lecturer taught you as right.

Do not simply accept it, and do not simply dismiss it. Both reactions waste the disagreement, which is the most informative thing that can happen in a practice session.

Work through it in this order:

Check whether you actually said it. Most of the time the point is in your head but not on the page — a half-sentence that assumed the reader would join the dots. That is not a grading error; it is exactly the mark you would have lost in the real exam, where the examiner also cannot read your mind.

Check the source. If it is a factual dispute — a date, an article, a threshold — go to the primary material. AI is confidently wrong often enough that this is worth the two minutes, and you will remember whichever version you had to look up.

Check whose framework it is using. Courses differ. If your lecturer teaches a four-part test and the model applies a three-part one, the model is not wrong in general but it is wrong for your exam. Grading against your own uploaded material reduces this, but does not eliminate it.

If it is still a genuine disagreement, keep it. Write it down and take it to a human. A specific question — "the grader says X is required here, my notes say it is not, which applies in this course?" — gets a real answer in thirty seconds of a lecturer's time. That is a far better use of office hours than handing over an essay and asking for thoughts.

The students who improve fastest treat disagreements as the interesting part. The ones who plateau either accept everything or argue with everything, and in both cases stop thinking.

The honest summary

AI does not mark better than your professor. It marks sooner, and it will do it ten times tonight.

For anything that counts — coursework, real assessments — you want the human. For the daily volume of attempts that nobody has ever had time to mark, and which is where the improvement actually comes from, a minute-long loop beats a week-long one by a margin that is hard to overstate.

The student who wrote one essay a week wrote eight before the exam. The one closing the loop in a minute wrote eighty.

Close the loop tonight: write an answer and have it graded against your own material, or generate the practice questions first. For what AI grading can and cannot judge in detail, see grading written exams with AI.