For teachers

A rubric for student thinking with AI

Andrew Mowat · MetaLearn Education

Most rubrics mark the finished product. That made sense when the product was good evidence of the thinking behind it. With AI, it isn't. A finished essay looks the same whether the student did the thinking or not, and the usual fix, an AI detector, doesn't hold up. One study found detectors flagged more than half of essays written by non-native English speakers as AI-generated (Liang et al., 2023). In an international school, that's a lot of students wrongly flagged.

What you can mark is the process: what a student checked, what they rejected, what they chose and why. This rubric does that.

How it works

Keep marking the product the way you do now. Add these rows for the thinking behind it, and give them 10 to 30% of the grade. Each row matches a zone of the Thinking Pendulum, and each asks for a piece of evidence the student hands in alongside the work.

The rubric

Zone	The task asks	Evidence	Strong	Developing	Limited
Mode 1	Declare what AI did	AI declaration	Says what AI did and why AI was right for that step. Notes or a plan show the thinking came first.	Says what AI did, but not why.	No declaration, or AI use only comes to light later.
Sense checking	Check claims; reject one	Keep / reject log	Checks claims against the text or a source. Rejects at least one AI response, with a reason, and corrects it.	Log is complete, but everything is kept or reasons are missing.	No log, or AI answers accepted unchanged.
Outcome determining	Three arguments first	Options page	Three different options. The choice is explained against the other two.	Options listed, but they're close to the same, or the choice has no reason.	One option only, or AI made the choice.
Problem shaping	Reframe the question	Question reframe	Reframes the question in their own words and finds a tension in it.	Restates the question with little change.	No reframe, or the reframe came from AI.
Whole journey	Share the link to each AI chat	Chat bibliography	Every AI chat is linked. The record matches the rows above, and you can see the thinking move between zones.	Some chats linked, with gaps in the record.	No chat links or transcripts.

Preview of the printable A4 Thinking Pendulum process rubric

Get the printable A4 version free with the guide.

Earned press or empty press?

A quick AI answer isn't a problem in itself. It's an earned press when the thinking happened earlier and the notes or plan show it. It's an empty press when there's no thinking behind it anywhere. From the outside, both look the same: a student typed a request and got an answer. That's why the rubric asks for evidence from outside the AI session. It scores the evidence, not the press.

Using it this term

Start with the Mode 1 and Sense checking rows. They're the easiest for students to understand and for you to see. Add the other rows as students get used to them. Ask for the chat bibliography from the start, because it lets you check the other rows against the real record. Use school accounts where you can, and accept an exported transcript where sharing isn't possible.

Questions teachers ask

Does this replace my marking scheme?

No. You keep marking the product as you do now. These rows sit alongside it and carry 10 to 30% of the grade.

How much time does it add?

About the same marking overall. The time is spread across the log and the essay, and the log often tells you quickly where the thinking happened.

What if a student doesn't use AI at all?

The rows still work. A keep or reject log, three options or a reframed question can come from books, notes or a conversation with a classmate. The rubric is about the thinking, whatever the tool.

Cover of the free guide: The Thinking Pendulum, a teacher's guide

Get the printable version

The printable A4 rubric comes free with the Thinking Pendulum guide, along with the model, five steps you can try this term and a planning prompt that builds the stages around your own task.

Get the free guide and rubric

The Pendulum in Practice editions go further, with a fuller rubric that tracks progress across year groups.

Further reading: Cognitive offloading & AI: what teachers can do

Further reading: AI as a thinking partner for students

References

Liang, W., Yuksekgonul, M., Mao, Y., Wu, E., & Zou, J. (2023). GPT detectors are biased against non-native English writers. Patterns. arxiv.org/abs/2304.02819