Most teachers who try AI grading once and quit do the same thing: they paste an essay in and ask for feedback. Without a rubric, the AI gives generic, syrupy comments that don't match how you actually grade. This kit fixes that. It is the workflow we see working in real classrooms, broken into four passes you can run on a stack of 28 essays in under 90 minutes.
Step 1: Anchor the AI to your rubric
Before you upload a single student paper, upload your rubric. TeacherAI's Grading tool stores it, parses each criterion, and uses it to anchor every comment to a specific row. If you don't have a rubric you trust, start with one from the standards library inside the tool and edit it. A 4-criterion rubric is enough to begin.
- Rubric uploaded with clear criterion names
- Each criterion has 3 to 4 performance levels
- Level descriptors use observable language, not vague adjectives
- Sample exemplar attached for the top level (optional but helps calibration)
Step 2: Calibrate on three papers, by criterion
Pick three papers spanning your expected range: a strong one, a middle one, a struggling one. Grade only the first criterion across all three. Compare the AI's score and comment to yours. If they drift, edit the criterion descriptor and rerun. You are teaching the AI your standard, in writing, once.
Why criterion-by-criterion
Reading one criterion across 28 papers is faster than reading 28 whole papers, and your judgement stays calibrated. The AI is the same way.
Step 3: Approve, don't read line-by-line
Run the full stack. TeacherAI shows the AI's score next to the sentence it came from. Your job is triage:
- Bottom 10 percent: read carefully, override where needed
- Top 10 percent: read carefully, the AI tends to under-celebrate excellent work
- Middle 80 percent: skim, approve, move on
Approval is a single keystroke. The cognitive cost of reading 28 mediocre essays in full is what burned you out; the cost of approving 22 AI drafts is nearly zero.
Step 4: Send same-day
Same-day feedback roughly doubles its impact. The point of grading faster is not the saved hours; it is closing the feedback loop while the writing is still alive in students' heads. Schedule the return for the next class block, not the next week.
Bonus: parent-communication scripts
When a grade is going to surprise a parent, send a short note before they see the gradebook. Three templates we hand out, each editable inside the AI Assistants tool:
- Heads-up before a low grade: warm tone, one specific cause, one specific next step.
- Pattern note over three assignments: data first, interpretation second, request for partnership third.
- Celebration note: specific to the work, not the student's character.
AI didn't replace my judgement. It replaced the four hours between my judgement and my students reading it.
Common pitfalls
- Skipping calibration. The AI will be 70 percent right out of the box; calibration takes it to 90.
- Editing every comment. If you are rewriting more than two in ten, your rubric is the problem.
- Returning grades in a batch a week later. The biggest gain from faster grading is faster return.