Grade the reasoning, not the artefact
If a deliverable can be generated in a minute, stop paying full price for it. Move the marks to what can't be outsourced.
Jul 23, 2026 · 1 min read
When the essay, the code, or the poster can be produced by a model, the honest response isn't better detection. It's shifting credit toward the parts of the work that still require the student: the decisions, the revisions, and the ability to explain them. This is the assessment-side answer to the OECD's finding that outsourcing to GenAI lifts performance without producing learning (OECD, 2026).
In practice
- Rebalance rubrics: fewer marks for the polished product, more for the plan, the draft-to-final changes, and a short spoken walkthrough.
- Ask for a "decision log": three sentences on what they changed and why. Fabricating a good one requires understanding the work, which is the point.
- Sixty-second explanations (see "Make them say it out loud") slot in here as the assessment layer, not an add-on.
- Say the quiet part to students: "I assume tools exist. I'm grading what you did that a tool can't."
Quick takeaway: don't defend the artefact. Devalue it.