Make every wrong option confess a misconception

Most multiple-choice questions tell you who got it wrong but not why. Wiliam (2011) argues that the value of a diagnostic question sits in its wrong answers: when each distractor is built from a known misconception, the spread of responses tells you which repair each group of students needs. AI drafts distractors quickly but defaults to plausible-sounding filler, so the prompt has to demand that every option maps to a named misconception. Deploy the best one as your hinge (see "Plan one hinge question per lesson"). (Manning's formative-assessment-loop-designer skill builds whole assessment loops on this principle.)

In practice

Write 1 multiple-choice question checking [objective]
for [year group].
Constraints:
- Exactly one correct answer.
- Every distractor must be the answer produced by a
  specific misconception. Name the misconception
  next to each option.
- No throwaway options: if you cannot name the
  misconception, replace the option.
Then add one line per distractor: what I should do
next with students who chose it.

• List the misconceptions you already meet in marking, then tell the AI each distractor must be the answer a student holding one of them would give.
• Reject any option that is merely wrong; if no misconception produces it, no information comes back when a student picks it.
• Ask for a one-line teaching response per distractor, so the reteach for everyone who chose B is planned before the quiz runs.
• Run it as a whole-class vote on mini-whiteboards or a quick poll; the spread across the options is the diagnosis.