The essay came back clean. You have no idea who wrote it.
Stop trying to police the artifact. Grade the trail of decisions the student made while working with AI, which is the thing you have been trying to measure all along and the one thing AI cannot hand them in a single prompt.
Yours forever. Delivered the second you pay.
You graded one standalone artifact and called it learning.
You do not know what they read. You do not know what they thought first and threw out. You do not know which sentences came from a flash of insight at 2pm and which came straight from ChatGPT at midnight.
That was always the deal, and before AI you could live with it, because the artifact at least came from the student. Now the artifact can be completely decoupled from the thinking.
Detection software loses. Better assignment instructions lose. You cannot out-design the fact that AI will produce any output you ask a student to produce.
Grade the trail instead.
An audit trail is the record of decisions a student makes while working with AI. The prompts they wrote, the outputs they kept, the outputs they killed, and the reason for each choice, captured while the choice is still live.
A student can ask AI to fake the trail too. A low-effort fake is easy to catch, and a high-effort fake costs roughly what doing the work costs. That is the whole trick.
One entry from a student trail
- Decision
- Modified
- Reason
- The ranking assumed isolation is only social. I reframed my prompt to separate social isolation from academic isolation, because my real interest is students who fall behind, not students who feel lonely.
That paragraph is the assignment. AI did not produce it, and no rubric you use today would have caught it.
The whole loop, in four steps
- 1
Add the four-field trail to one assignment you already grade.
- 2
Tell students the trail is worth more than the artifact.
- 3
Score the trail against six observable rules, under two minutes per student.
- 4
Require a What Changed note, and run the Monday starter if you want a clean first try.
Everything is copy-paste. You can run the first piece in your next class without rewriting a single assignment.
The seven sections
Each one is paste-ready. The prompts are written out in full, with the bracketed inputs marked.
The Audit Trail Format
The four-field entry students fill in every time they make a real decision with AI: the prompt, the output, the call they made, and why. Paste-ready handout with a worked example.
The Assignment Wrapper
Language that makes the trail the graded object rather than extra credit, plus a prompt that converts an assignment you already use without changing your learning goal.
The Six-Rule Rubric
Six observable thresholds you score instead of judging polish. Thickness floor, rejection ratio, prompt evolution, reason quality, dissent, and the override bonus.
Spotting a Faked Trail
The four tells of a fabricated trail, and a prompt that flags one for your review without making an accusation. You keep the final call.
The What Changed Note
A 150-word note contrasting the Day 1 belief with the Day 7 belief. The single hardest thing in the system to fake.
Team Trails
Three extra fields that expose who thought and who coasted: whose decision it was, what dissent got logged, and what resolved it.
The Assignment You Can Run Monday
A complete starter assignment, already wrapped, so your first try does not have to be a rebuild of something you care about.
What this will not do
It will not catch cheaters. Detection is the game you were already losing. This changes what you ask for so that the thing worth grading is the thing that got produced.
It will not survive being bolted on as extra credit. If the trail is worth 10 percent, students will treat it like it is worth 10 percent, and you will have added busywork to both of your lives.
And you do not have to convert your whole course. Take one assignment you already grade and run it once this semester.
Seven sections, every prompt written out, the student handout, the six-rule rubric, and a starter assignment you can run Monday. Yours forever, and it works on every course you teach from here on.
Get Stop Grading FumesSecure checkout through Stripe. You get it immediately after payment.
This is also included in the How To Teach With AI membership, along with every other system and tool I have built, for $40 a month.
See the membershipThe stack is coming either way
In December you will sit down with forty clean, competent, interchangeable papers and try to guess which students learned anything. You will guess wrong, and you will know you guessed wrong, and you will submit the grades anyway.
Or one assignment this semester comes back with the thinking attached, and you grade what actually happened.
Nobody is asking you to rebuild your course. One assignment. You already know which one.
Get it for $47