Let AI mark against your rubric, then disagree with it
Not to save marking time. Because defending your mark against a confident second opinion is the fastest rubric-sharpening exercise there is.
Sep 1, 2026 · 1 min read
Give an AI your rubric and three anonymised pieces of work, and it will produce scores with reasons. The value isn't outsourcing the judgement; the OECD's review of GenAI in classrooms is consistent that teacher agency has to stay in the loop (OECD, 2026). The value is moderation on demand: everywhere the AI's score differs from yours, either your rubric wording is loose or your marking drifted, and both are worth knowing before you mark the other twenty-seven.
In practice
- Mark the three pieces yourself first, then compare. The order matters: AI-first anchors you to its numbers.
- Interrogate every disagreement: "Which rubric line justifies your score on criterion 2?" If it can cite your own wording against you, tighten the wording.
- Anonymise the work first, always (see "Never paste student names into a chatbot"), and the recorded mark stays yours.
Quick takeaway: use the AI as a moderation partner, not a marker. The disagreements are the product.