Output-diff critique

Compare two AI drafts and say which is better and why: training criteria, not passive acceptance of the latest regenerate.

When to use this: When checking yourself
Learning impact: Clear help, modest gain
Learning evidence: Thin or mixed evidence
Where you are with the topic: Checking readiness
How often to use it: Worth repeating often

When to use it

You have two AI drafts and were about to keep the longer one by default.
You need to practise evaluative criteria, not passive acceptance of the latest regenerate.
A coursework draft needs a merge you write yourself.

What to ask AI

Type something like: “Draft A: [ ]. Draft B: [ ]. Do not declare a winner. Ask me to rank and state criteria.”

Then have it attack inconsistent criteria.

Keep the hard part on your side: you produce the merged version.

How it helps

Comparative judgment trains criteria more than reading a single polished draft.

Diffing two AI outputs is that practice on tool text.

The risk is aesthetic preference: longer, denser, or more confident wins without reason.

Force criteria first, then choose, then merge in your own words.

How to run it

Generate or collect two drafts for the same brief.
Before reading preferences aloud, write your criteria: accuracy, structure, audience fit, evidence.
Rank with reasons. Refuse a winner declared by the model.
Let it probe contradictions in your criteria.
Merge yourself. Do not ask it to autogenerate the final as a third mystery draft.

Example exchange

You

I’ve attached Draft A and Draft B. Please don’t declare a winner. Ask me to rank them with clear criteria first.

What matters more for this assignment: mechanistic accuracy, clarity for a peer, or citation density?

You

Accuracy first. A wins on mechanism; B wins on clarity. I favoured B at first just because it was longer.

Your criteria just caught an aesthetic bias. When you merge, keep A’s mechanism and B’s signposting, and write the merge yourself.

Copyable prompt

Draft A: [ ]. Draft B: [ ].
Do not declare a winner. Ask me to rank and state criteria. Then attack
inconsistent criteria. I will produce a merged version myself.

The Tell

Here is how you know this method has flipped: you always keep the longer draft.

Length became your only criterion.

Write criteria before you look, then merge on purpose.

Principle evidence

Strength of the underlying learning idea, not a claim about AI products.

The underlying learning idea is rated weak-to-moderate. Comparative judgment and peer-assessment-adjacent skills train evaluative criteria more than passive acceptance of a single draft. Diffing two AI outputs is that practice on tool text. AI-output tournaments as controlled pedagogy remain speculative.

AI delivery evidence

Whether an AI tutor delivers this method well is a separate question.

Models can help probe your criteria. They should not silently pick winners. Delivery speculative; your stated criteria and self-written merge are the learning.