How to mark essays against a rubric — a teacher's guide (and where AI fits)

If you have ever sat down to mark a stack of essays against a rubricat nine o'clock on a Sunday night, you already know the quiet problem with it: the rubric is meant to make marking consistent, but holding six criteria in your head across thirty scripts is genuinely hard. By essay twelve you are marking against a slightly different internal standard than you were at essay two. This is a teacher's guide to marking against a rubric so that the last script gets the same judgement as the first — and an honest look at where AI can carry some of the load without taking your professional judgement out of the loop.

What a rubric actually is (and isn't)

A rubric is an instrument, not a decoration stapled to the back of the task sheet. Each criterion names an observable quality of the work — "sustains a clear argument", "analyses rather than describes", "selects and integrates evidence" — and each band describes what that quality looks like at different levels. The marker's job is not to form an overall impression and then find a number to match it. It is the reverse: judge each criterion against its descriptor, then let the marks add up to the grade.

That distinction matters because it is where reliability comes from. When two markers moderate a script and disagree, they almost never disagree about the whole essay. They disagree about one criterion — was that analysis, or a well-written description? Marking criterion by criterion turns a vague argument about quality into a specific, resolvable question. It is the same discipline students benefit from when they learn to read a marking guideline like a marker.

How to mark an essay against a rubric, step by step

The method below is nothing exotic — it is what experienced markers do by instinct, written down so it survives essay number thirty.

• Read the whole script once, pen down.Resist marking on the first pass. You are getting the shape of the argument so that a strong conclusion doesn't get punished for a slow opening, and a slick opening doesn't flatter a hollow middle.
• Take one criterion at a time. On the second pass, hold a single criterion in mind and ask only: which band does this script sit in for this quality? Underline the evidence as you go — the actual line that shows the analysis, or its absence.
• Anchor to the descriptor's verbs. The words that separate the top band from the middle — skilfully, perceptively, sustained versus soundly, describes, some — are the whole game. Decide which verb honestly describes what is on the page.
• Mark the criterion, then move on. Record the band before you read the next criterion. This stops a brilliant first criterion from bleeding upward into the rest of the marks.
• Only then, total and sanity-check.Add the criteria, look at the grade, and ask whether it feels right. If it doesn't, the fix is to revisit a specific criterion — not to nudge the whole number.

Two habits keep a class set honest. Keep an anchor script or two beside you — one you have already marked and are confident in — and glance back when your standard starts to drift. And mark one criterion across several scripts in a row when you can; you calibrate faster judging the same quality repeatedly than judging six different qualities on one essay before starting again.

The comment problem: turning marks into feedback

A rubric produces a mark efficiently. It does not, on its own, produce the sentence a student can act on. "Criterion 3: developing" is accurate and nearly useless to a fifteen-year-old. The comment that moves a student is specific and located: "You quote the evidence in paragraph two but then restate the plot instead of analysing the technique — that's the difference between this band and the next one."

Writing thirty of those is where the evening disappears — especially when eight students have made more or less the same mistake and you are typing a variation of the same paragraph eight times. This is the real cost of rubric marking, and it is worth naming: the marking itself is quick; the individualised, criterion-linked comment is slow.

Where AI fits — honestly

This is the seam an AI marker can help with, and it is worth being precise about what it does and doesn't do. Mark takes the rubric you upload and the essay, and returns a first-pass mark for each criterion with the quoted line it used as evidence — the same criterion-by-criterion shape you would apply yourself, in about a minute per script. It does not write or rewrite anything for the student, and it does not have the final word: every score is yours to confirm or overrule, and you remain the marker of record.

Used that way, the tool changes marking from writing into reviewing. You are no longer generating the judgement from a blank page for the thirtieth time; you are checking a proposed judgement against your own, agreeing or correcting, and spending your saved time on the comments that actually need a teacher — the borderline scripts, the unusual arguments, the student you know is capable of more. The consistency is a side benefit: the same rubric is applied identically to script one and script thirty, because it isn't being held in a tired human memory.

A word on integrity, because students watch what we model. Using AI to help mark — where the teacher confirms every result — is a world away from a student using unapproved AI to produce assessable work, which in NSW is treated by NESA as malpractice. If your students use an AI marker on their own drafts to prepare, that is feedback, not submission — and it is worth teaching them to acknowledge any AI feedback honestly so the line stays clear.

The short version

Marking against a rubric well is a discipline: read first, judge one criterion at a time, anchor to the descriptor's verbs, and total last. The bottleneck isn't the marking — it is the criterion-linked comment, written thirty times over. That is exactly the part worth sharing with a tool that proposes a mark per criterion for you to confirm, so your judgement goes where it is needed most and every student in the class is measured against the same standard. Teachers get 25 free credits to try it on a real set; see how it handles your rubric before you trust it with a stack.