K16

Transcript review

Measure it, by reading your own transcript back.

When to use this: At the end of a session
Learning impact: Reliable gain
Learning evidence: Credible with caveats
Where you are with the topic: Checking readiness
How often to use it: Worth repeating often

When to use it

A session felt great and you cannot say how much of the thinking was yours.
You want evidence of ownership, not a warm feeling of progress.
You are wrapping a long chat and suspect you mostly approved explanations.

What to ask AI

Type something like: “Here’s the transcript of my last study session. Don’t summarise it. Tell me what proportion of the substantive ideas I generated versus you.”

Ask where you accepted something without testing it, and what you could reproduce tomorrow unaided.

Keep the hard part on your side: you judge the labels; it only helps you notice patterns.

How it helps

Agreement is nearly indistinguishable from comprehension in the moment.

Following a good explanation feels like generating one.

Transcript review measures who did the reasoning by reading the record back.

Mark each substantial idea: you produced it, or it produced it. Count.

The discomfort is the point.

How to run it

Export or paste the transcript without asking for a glowing summary.
Mark who asked the questions. Consumer sessions are mostly you asking and it answering.
Find each “click” moment and read the three turns before it. Handed or earned?
Compare message length. Three-word replies against four-paragraph answers usually means you were reading.
Note whether you ever disagreed. No friction often means no evaluation.
List what you could reproduce tomorrow unaided. That list is usually the shortest and the most honest.

Example exchange

You

Here’s last night’s transcript. Don’t summarise it, tell me roughly what share of the substantive ideas I generated versus you.

Roughly twenty percent from you, eighty from me. You also accepted an “always” claim about enzyme denaturation without asking for a temperature range.

You

Okay. Quiz me on what I should still be able to say without looking.

Unaided: what happens to reaction rate as temperature rises, before denaturation sets in?

Copyable prompt

Here's the transcript of my last study session. Don't summarise it. Tell me:
what proportion of the substantive ideas did I generate versus you; where did I
accept something without testing it; and how long were my contributions relative
to yours?

The Tell

Here is how you know this method has flipped: you’ve never audited a session.

Then you don’t know, and the feeling of a good session is exactly what you cannot use as evidence.

Audit one transcript this week before you trust that feeling again.

Principle evidence

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

The underlying learning idea is rated moderate. Reflective review and metacognitive monitoring of help-seeking reveal whether the learner generated ideas or only approved them. Principle moderate for post-session ownership checks; AI transcript audits as delivery inherit the reflection construct but remain speculative as products.

AI delivery evidence

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

A model can help tally patterns in a paste. It can also flatter you. Delivery speculative. Your honest labelling and unaided quiz are the method.