Generate student exemplars at every grade band

Generate student exemplars at every grade band

Students who have never seen an A-grade response cannot write one. Ron Berger has been saying this for thirty years, the assessment literature has piled in behind him (Wiliam on comparative judgement, the EEF's feedback guidance, the Making Good Progress work), and every teacher who has ever run "here are four responses to the same task, rank them before you write yours" knows it lands. The reason the move is rare, not the reason it does not work, is that writing four plausible student responses at four different bands to the same task is real work. Ninety minutes of it, easily. This is the classic case of AI collapsing a genuinely useful but genuinely slow lesson-prep move into ninety seconds.

The prompt

You are helping me write four exemplar student responses to the same task, one at each grade band. I will use them tomorrow in a critique lesson where students compare the four side by side and annotate the differences.

You are working in the tradition of Ron Berger (An Ethic of Excellence, models and critique), Dylan Wiliam (comparative judgement, success criteria) and the Australian assessment literature on standards-referenced grading. That means:
- Every exemplar must be a plausible response a real student at that band would actually write, including the specific errors, gaps or brilliance that band demonstrates.
- All four exemplars must address the same prompt so the differences between bands are legible.
- The differences must map to concrete features (structure, use of evidence, precision of language, depth of analysis), not vague "quality".
- The lowest-band exemplar is NOT joke-bad. It is what a student who came to the lesson and tried, but did not have the skills or knowledge yet, would produce.

THE TASK

Year level: [YEAR LEVEL, e.g. Year 10]
Subject: [SUBJECT, e.g. English]
Task type: [TASK TYPE, e.g. 400-word text response paragraph, mathematical proof, historical source analysis, scientific method report]
The exact task or question I set the students:
[PASTE THE TASK VERBATIM]

THE RUBRIC / SUCCESS CRITERIA
[PASTE THE RUBRIC IF YOU HAVE ONE. If not, describe in three or four bullet points what an A response looks like: what it contains, what it argues, how it uses evidence, how long it is. The model needs this or it will invent a rubric of its own.]

THE BANDS

Produce four exemplars at the bands my rubric uses. If my rubric uses A/B/C/D, use those. If it uses Very High / High / Medium / Low, use those. If it uses 1 to 5, pick 5, 4, 3, 2. Ask me once if it is not obvious from the rubric I pasted above.

OUTPUT

For each of the four bands, produce three things in this order:

1. THE EXEMPLAR. The full student response, at the length a real student would write it. No stage directions. No "note the use of...". Just the response, as if handed in.

2. WHY THIS IS A [BAND]. Three or four bullet points, each pointing at a SPECIFIC feature of the exemplar above (a sentence, a paragraph, a step) and naming what it does and why it lands at this band. Reference the rubric directly.

3. THE ONE MOVE TO CLIMB. What is the single change that would move this exemplar up one band? Specific, actionable, something a student could actually do in a five-minute rewrite.

THEN produce THE LESSON MOVE. Four to six lines. How I run this in a 45-minute lesson: how students compare the four exemplars, what they annotate, what they do next. Assume a mixed-ability class of 25 in an Australian [PRIMARY / SECONDARY / TERTIARY] school.

CONSTRAINTS
- Australian English (analyse, colour, realise).
- No em-dashes or en-dashes.
- No preamble. Start with the top-band exemplar.

What it does well

Produces four full exemplars of genuinely usable quality in about ninety seconds, plus the per-band annotations that turn a stack of exemplars into a teachable lesson. Most teachers who try this by hand stop at "here are four responses" and never get to the annotation, which is where the learning actually is. This prompt closes that gap for you.

Where it falls over

The lowest-band exemplar can drift into caricature if you skip the rubric section. Feed it the actual success criteria, or the model invents what "not good" looks like from stereotype and you end up with an exemplar no real Year 8 would ever produce. The top-band exemplar is also worth reading yourself before showing the class. On rare tasks it will land more polished than a real student in your year level could plausibly write, which teaches the class the wrong ceiling. Trim it back if it does.

Try it like this

For Year 8 Humanities: paste your source-analysis question and the rubric, generate four responses, and run "which is the B response and why" as the starter next lesson before students write their own.

For VCE English: draft four introduction paragraphs at each grade band for the SAC prompt, hand them out unlabelled, and have the class rank them before revealing the bands and drafting their own.

If you only have ten minutes: skip the lesson move, print the B and the A exemplars, and run "add one sentence to the B to turn it into an A" as an exit ticket.

The prompt takes ninety seconds. Reading the four exemplars back with a red pen before you print them is the ten minutes that separates using AI well from using it credulously.