What 'Acceptable AI Percentage' Actually Means on a Rubric

Course policies often quote a maximum AI percentage, and the quotation hides a problem: the percentage is a model estimate, not a measurement of how many words a tool wrote. Two detectors will disagree about the same essay, and a slightly different version of the same essay can swing ten points.

That does not make the policy meaningless. It makes the wording the important part. Some policies care about disclosure, some about the final voice being the student's, some about process evidence. Those are answerable questions. "Was the score below 20" is not, because the score moves.

The practical response is to keep the work defensible: your outline, your notes, one draft before any tool touched it. Then the percentage is a reading of your prose, not a verdict on your process.

This page on acceptable AI percentages explains how institutions use the thresholds, and this piece on detector accuracy explains why the numbers move.