Give AI the final exam before the first lesson

Backwards design starts from the evidence you will accept, not the activities you enjoy teaching (Wiggins & McTighe, 2005). Most AI planning prompts run the wrong way: teachers ask for lessons on a topic, get plausible activities back, and discover at marking time that the sequence never rehearsed what the assessment actually demanded. Reverse the order of the conversation. Give the model the final assessment first and make every proposed lesson justify itself against that evidence. (Manning's backwards-design-unit-planner skill formalises the full three-stage version.)

In practice

Here is the end-of-unit assessment and mark scheme:
[paste assessment]
Unit length: [n] lessons. Year group: [year].

Plan backwards from this evidence:
1. What must students know and be able to do to
   succeed on each question?
2. Sequence [n] lessons building exactly that.
3. Flag any activity that does not serve the
   assessment, and any stated outcome the
   assessment fails to test.

• Write or choose the end-of-unit assessment first, with its mark scheme, and paste it in full; it is the clearest statement of intent you own.
• Ask for a lesson sequence where every activity earns its place by moving students toward that evidence, and make it flag anything that does not.
• Ask what the assessment fails to capture from your stated outcomes, then fix the assessment before you plan a single lesson.