Skip to content

Everyone Uses AI Until the Teacher Does

A French column caught students demanding that their teachers account for their own AI use. Under the hypocrisy sits a real distinction: a deliverable and a demonstration of skill are two different products, and AI is tolerated in only one of them.

AI-assisted draft. Reviewed and edited by the Phosphene team before publication.

On this page
A pale goose quill crossed over a matte black pen on a dark walnut desk, a blank envelope and a blurred ink bottle behind, lit by window light from the left.
Editorial illustration, not a document from the story.

Le Monde's weekend magazine ran a column by Guillemette Faure this week about a sentence French parents keep hearing from their teenagers: the teacher uses ChatGPT. The evidence is almost never a confession. It is formatting. Every paragraph the same length. Bullets. A few bolded words. Arrows, sometimes a smiley. A handful of clues in the layout of a handout, and the kid is standing in the kitchen, outraged.

Faure's best line lands the scene: the student delivers the accusation with the moral disgust of a Michelin inspector who has just found a Picard box in the kitchen of a starred chef. Everyone in that kitchen uses these tools. The kid, the parents, the neighbours. A teacher, somehow, is still expected to write the course out with a quill.

Then the inversion. The most indignant students, she writes, are often the ones whose ChatGPT history could reconstruct the whole school year. They are precisely the people who know how to ask a model to write like a ninth grader. Caught, they say it was just to check, and anyway not for everything. But the teacher, in their judgement, did not tire himself out.

So the students who are now required to prove their homework came from their own brain have started requesting the same proof from the people grading it. They watch the ones who watch them.

The hypocrisy is not the interesting part

It would be easy to stop at the double standard, and most commentary on this story will. The double standard is real but it is also boring, because it is available in every generation and every profession: the rule is strictest for whoever is not in the room.

The reason this particular accusation stings is structural. A teacher's handout is two products stapled together, and only one of them is about producing a course. The other is about producing evidence that a skill exists and has been transferred. AI is a legitimate tool for the first and a destructive one for the second, and the two are printed on the same sheet of paper. That is what the students noticed, even if they phrased it as a joke about a frozen meal.

Worth saying plainly: the mockery is not evidence that the course was bad. A teacher who builds a handout with a model and then teaches from it can be doing better work than one who retypes the same notes every year. The complaint is not about quality. It is about audit.

Your formatting is now a fingerprint

The detail that should worry anyone who publishes things is how the students knew. Not a watermark, not a tell in the prose. The shape of the page. Evenly sized paragraphs and a bulleted structure with bold lead-ins were neutral design choices three years ago. Now they read as evidence of a specific process, and the reader supplies the conclusion before you get a chance to explain.

Anyone who has looked at a wall of machine-made event posters knows the feeling from the other side. You clock the artifact in under a second, and then you decide something about the person who put it up. In our earlier look at the French poster backlash the anger was rarely about pixels. It was about a message the audience read as nobody here spent a minute on this.

The same reading now applies inside a classroom, a studio, a pitch, a portfolio review. Your audience has learned a new skill in the last two years, which is inferring process from surface. That skill does not switch off when it is aimed at you.

Two products, one file

If you sell creative work, the useful move is to decide which product you are making before you open the tool.

There is the artifact, and there is the demonstration. A bakery wants a poster that sells bread. Nobody at the bakery will ask how it was made, and if the poster works, the question never comes up. But an illustrator selling character design is not selling a picture. They are selling a claim about what they can do again next month, for a different brief, without luck. The picture is the proof of the claim. That is why the audit lands there and not at the bakery.

The confusion between the two causes most of the bad outcomes I see. People use AI to make a deliverable, then defend it as if it were a demonstration, which is indefensible and makes them look dishonest. Or they hand-make everything for a deliverable no client will ever inspect, and wonder why the week disappeared.

Ask the question in this form: is someone going to judge the thing, or judge me? A grant jury, an art school, a competition, a client hiring for a skill, a collaborator deciding whether to trust you with the next job — those are all judging you. A detector score is not a verdict, but the fact that people reach for one tells you what they are trying to establish. They want to know whether the person is present in the work.

The receipt that costs nothing

If you are in the second category, the practical question is what evidence to keep, and the honest answer is: less than you fear, but earlier than you think.

The working file is the strongest receipt you will ever have, and you already produce it. A layered document, a project file with editable text, a comp with the type still live. Nobody has to trust you about a file they can open and take apart. That single habit answers most accusations before they are made.

Next comes a decision log, which is cheaper than it sounds and better than a recording. Ten words per rejected take, with the reason. Wrong light. Hands broke at this angle. Too soft for print. A dated list of rejected options and why is far more convincing than a timelapse, because it is the part of the process that cannot be faked by performing motion.

Which brings up the thing to avoid: process theatre. A screen recording of someone dragging a slider proves that a hand moved. Everyone can produce that, including someone who generated the image in another window and is now moving a slider over it. Evidence is decisions, not keystrokes. If the only proof you can offer is motion, you are offering nothing that a nervous client could not fake themselves.

And match the disclosure to what is being judged. A client buying a deliverable wants to know what they are getting and whether they can use it, not a method statement. A jury, a teacher, an editor commissioning a byline wants the method. Telling the wrong audience the wrong thing reads as either evasion or oversharing.

The standard you enforce is the one you have to keep visible

The part of Faure's column that will outlast the handout jokes is about authority. Students were asked to prove their work came from their own heads, by people who were not visibly doing the same, and the students noticed the gap. That is not a school problem. It is what happens in any studio where a creative director asks for craft and then ships something obviously machine-made. The note stops landing. Not because the note was wrong, but because the person giving it is standing on the wrong side of their own rule.

The fix is not abstaining. It is keeping the standard you enforce somewhere a person can see it: in the working files, in the drafts you show, in how you talk about what the tool did and what you did. The students in that column will forget the handout. What they will remember is discovering that the rule had two versions.

Assume the audit is coming for you too. Decide now which version of the rule you are on, because the audience has already learned how to check.

Sources

Share this guide
X LinkedIn