I want to allow AI and still see thinking

What is going on

You do not want to police, and you do not want to pretend AI is not there. The difficulty is that a finished answer no longer shows who did the thinking. So "AI allowed" with nothing else tends to mean you see less and less of your students' own work.

The fix is about order. In a classroom study of 42 undergraduates (Islam and others, accepted at AIED 2026, a conference paper), asking students to reflect first and then consult the AI gave the largest drop in over-trusting AI answers. It is a small study and it measured trust, not learning. It supports the common-sense design: thinking first, AI second.

What to do this week

1. Say where AI is allowed

   One sentence on the task sheet: "You may use AI after your written attempt, to get hints and to check your work. You may not use it to write the answer."

2. Collect the attempt

   Five to ten minutes, in class or at home. A guess, a plan, a first paragraph, or a clear note of where they are stuck. Credit for real, not for right.

3. Teach one prompt

   Show the class how to ask for coaching: paste the attempt, then "Don't give me the answer. Guide me with questions, one at a time." Five minutes of modelling saves weeks of misuse.

4. Collect "what changed"

   Two lines: what I changed after getting help, and why. This is the part that is hard to outsource without it showing.

What not to do

• Do not allow it everywhere at once. "Use AI responsibly" gives students nothing to hold on to.
• Do not ask for chat logs as routine. They are long, easy to stage and private. The attempt tells you more.
• Do not point students at a tool they are too young for. Check your school's list first.

The honest limit

A student can write the attempt after the answer. Some will. In-class attempts close most of that gap, and the "what changed" lines are harder to fake than they look. This shows you thinking. It does not prove who did it.

Questions