I handed back a stack of essays with a scored rubric attached to each one, and within a week half of them were in the recycling bin, ungraded feedback and all. The rubric had done its job for me, it justified a grade, but it hadn't done anything for the student, because a column of checked boxes doesn't tell you what to do differently next time.
The rubric's actual job, if it has two
A rubric scores work. That's one job. The second job, the one most rubrics skip, is telling a student exactly what separates their score from the one above it. If your rubric can't answer "what would have made this a level 4 instead of a level 3," it's only doing half its job.
What makes the difference actionable
- Concrete, observable language per level, not comparative adjectives. "Uses two pieces of textual evidence" beats "uses some evidence" every time, because a student can count to two.
- The gap between levels stated in the same terms. If level 3 is "uses one piece of evidence" and level 4 is "well-supported," those aren't comparable, the student can't see the actual gap.
- Criteria a student would use themselves before submitting, not just ones you use to grade after. If a criterion is only useful to you, it's not helping them improve.
A prompt that gets you a genuinely usable rubric
"Create an analytic rubric for [assignment]. Criteria: [list your 3-4 main criteria]. 4 performance levels. For each criterion, write each level using specific, countable, observable language, not vague adjectives like 'good' or 'excellent.' The difference between adjacent levels should be something a student could check for themselves before submitting." Read what comes back and test it: could a student, reading only the rubric, tell you what to add to move up one level? If not, it needs another pass.
Where AI genuinely helps and where it doesn't
Helps: turning your vague first-draft criteria ("good analysis") into the specific, countable language above, fast, and generating multiple performance-level phrasings so you can pick the clearest one.
Doesn't help: knowing which criteria actually matter for your specific assignment and your specific students, that's your call, informed by what you've seen go wrong in past submissions, not something a general prompt can guess.
Using the rubric before the assignment, not just after
The single biggest change I made wasn't the rubric's wording, it was handing it out on the day I assigned the essay, not the day I returned it. Students who read the rubric before writing ask better questions and make fewer of the mistakes that cost them a level. A rubric only used for grading is a scoring tool. A rubric handed out at the start is a writing tool.
A quick example of the shift
Vague level 3/4 split: "Some analysis" vs. "Strong analysis." A student can't act on that. Specific split: "Identifies the central idea and connects it to one piece of evidence" (level 3) vs. "Identifies the central idea and traces its development across at least two moments in the text" (level 4). Now a student revising their draft knows exactly what to add.
One rubric, reused all year
Once you've built a genuinely specific rubric for a skill, like "textual evidence" or "claim and counterclaim", reuse the same language across assignments all year. Students stop treating each rubric as a new mystery and start recognizing the criteria as the same skill, assessed again, which is exactly what builds transfer.
Our [grading rubric generator](/grading-rubric) builds the analytic rubric above from your assignment and criteria, with specific, leveled language for each performance tier, ready to hand out before the assignment, not just after.
