The Work Before the Label

You read a page and feel something. A sentence catches a thought you had never quite managed to express. You read it again. You send it to someone.

Then you hear that AI may have helped make it.

What, exactly, has changed?

The words remain where they were. Your first response happened. New information about the work’s origin may give you a good reason to revise part of your judgment. It may raise a question about credit, honesty or the kind of achievement being celebrated. But it cannot travel backward and prevent you from having been moved.

There is something intellectually weak about a culture that answers every one of those questions with a single act of rejection. The possibility of AI involvement becomes a stain. It is enough to say the word, and the work must be thrown away.

I think we should resist that reflex. We should be curious about what a tool can help people achieve. We should also be precise about what we are praising, what was disclosed and whose work may have been used. A serious culture can do more than one kind of judging at a time.

This matters beyond books and pictures. The same reflex can enter arguments about research, education and public service. Once a technology becomes a symbol of everything people fear, evidence of its usefulness may be treated as a moral inconvenience. A person who admits to being helped is asked to apologize for the source of the help.

That is a poor way to discover what the future can offer.

The case that needs all its facts

On September 21, 2026, Thélyson Orélien’s C’était ça ou mourir won the Prix du Roman Fnac. Four days later, the Académie Goncourt removed the novel from its longlist amid allegations of AI generation and plagiarism involving earlier work.¹²

Orélien denied using AI to write the novel. His Canadian publisher, Les Éditions du Boréal, said it could neither confirm nor refute the AI allegations at that stage. The publisher separately expressed concern about plagiarism allegations involving other texts.³⁴

These distinctions are essential. This was a Fnac winner removed from the Goncourt selection, not an AI-authored novel that had won the Goncourt. The AI claim remained disputed. The controversy involved questions beyond the use of a tool.

It would therefore be dishonest to use the episode as a clean experiment proving that people loved a book until they learned its true method of composition. The method had not been established by the public evidence described here. And plagiarism is a different issue from assistance.

The case does expose the urgency of getting our categories right. A prize decision can involve eligibility, trust, the conduct of an author and a judgment of the text. Public discussion can compress all of those into a headline about AI. Once compressed, the story becomes available to whichever side wants a simple lesson.

We should decline the shortcut.

A reader can admire a novel and still believe its author should answer a serious allegation. A prize can have rules about human authorship without settling the literary value of every ineligible work. A person can object to plagiarism while remaining open to disclosed AI assistance. These positions do not require confusion or hypocrisy.

What they require is the patience to identify the actual claim.

If the allegation concerns copying, examine the alleged copying. If it concerns a false declaration, examine the declaration and the evidence. If it concerns a competition rule, read the rule that applied. If the claim is that a paragraph is bad, show us what is bad about the paragraph.

The word AI should not be allowed to do the work of evidence.

When the label moves the verdict

There is cleaner evidence for an effect of origin labels. Manav Raj, Justin M. Berg and Rob Seamans reported 16 preregistered experiments involving 27,491 participants in a 2026 paper. Evaluations of creative writing fell when participants believed AI had generated or helped produce it.⁵

In one study, 1,572 U.S. participants evaluated selections from five prize-winning, human-written short stories. Labels describing the same texts as AI-generated or made by a human using AI reduced their appeal compared with human-only labels. The experiment shows that a belief about origin can change a judgment without a change in the text. It does not show that AI generally writes better than people, or settle the value of authorship.

The finding gives us a useful moment of discomfort. If we say a passage is moving, then call it empty after its label changes, we should ask which part of our response we are describing. Did we find a defect in the prose? Did we revise our estimate of the maker’s skill? Did we lose a feeling of connection with a person? Did we become concerned about the conditions under which the work was made?

All are possible. They are not the same judgment.

A photograph of a landscape and a photograph of a real rescue can be visually similar to staged images while carrying different meanings. A reader may value a memoir partly because the events happened to its author. A handmade object may be prized partly for the labor and skill embodied in it. Information about origin can properly affect value.

But the fact that origin can matter does not establish that any AI involvement cancels every other kind of value.

A novel is not usually a promise that the author never received a suggestion, used a reference or revised with help. The relevant promise may instead concern the author’s creative responsibility, the originality of the expression and the honesty of any stated method. The expectations of a particular publisher or prize may be narrower. We need to state those expectations rather than assume they are obvious and universal.

Otherwise, a spell-checker, a research assistant, a translator, a tool that generates a first draft and a system that produces an entire submitted work can be forced into the same box. That box is too crude to support a fair judgment.

The question is what the person and the system each did, and why that division matters to the claim being made.

The fear beneath the insult

Not every objection to AI art is a failure to appreciate a good picture. Some objections concern the livelihood and authority of the people who make culture.

An illustrator may fear that a client will use a tool to avoid paying for work. A writer may object to the use of copyrighted material in training. A translator may worry that skilled judgment is being treated as a final polish on cheap generated text. These are questions about markets, rights and power. A viewer’s pleasure in an image does not answer them.

Nor does the existence of those questions prove that every assisted work is worthless.

We should be able to say that a result is good while demanding fair conditions for producing it. We do this whenever the quality of a product and the treatment of its makers raise separate concerns. The pleasure of use should not silence the labor question. The labor question should not force us to lie about the quality of the result.

My objection is to the leap from a legitimate dispute to a doctrine of contamination: anything touched by the tool is beneath consideration. That doctrine saves its adherent the trouble of examining the work. It also blocks the possibility that new forms of authorship may produce things worth reading, hearing or seeing.

It can have another effect. It may reward concealment. If a minor use of a tool attracts the same condemnation as a major misrepresentation, a person has little reason to explain the difference. The public receives less useful information, while the most careful disclosures become the easiest targets.

A better culture would ask for disclosures that illuminate the making of the work. It would distinguish meaningful assistance from incidental use. It would assess a false claim as a false claim, rather than treating every admission as a confession of worthlessness.

This does not require a universal label on every sentence ever revised with software. Different settings need different information. A medical article, an examination, a commercial illustration, a personal poem and a literary prize have different purposes. The disclosure should help the relevant audience understand the work and the claim attached to it.

A rule that no one can explain will mostly be enforced through suspicion.

We should be especially wary of judging origin by a list of stylistic habits. People repeat words. People use neat structures. People write awkwardly, fluently, formally and plainly. A sentence that sounds familiar may reflect a convention, an editor, a language learner or a tired author. It is a poor basis for a confident allegation about how a work was made.

If authorship is important enough to govern a prize or a person’s reputation, the process for examining it should be better than a hunch about punctuation.

What are we asking the artist to do

There are several different achievements an audience may want to honor.

One is mastery of a technique. Another is the quality of a finished work. Another is the choice of an idea, the arrangement of material or the ability to make something speak to a particular moment. A single work can display several of these achievements at once.

New tools can change their relationship. When a difficult operation becomes easier, the skill required to perform that operation may become less rare. Other decisions may become more important. The person who chooses, edits, rejects and shapes can still exercise judgment, but we should examine the judgment rather than assume it exists whenever someone presses a button.

It is equally lazy to say that all prompting is artistry and that no work involving a model can contain artistry.

The interesting question is what was made possible and what the maker contributed. Did the person discover a form worth pursuing? Did they recognize when an output failed? Did they transform the material? Did they take responsibility for the facts, implications and people represented? The answers may differ greatly between two works made with the same tool.

A good work can also exceed the skill of its maker in a particular operation. That is one of the reasons people build tools. We should not automatically regard such assistance as a moral defect. A person who has an idea but lacks one technical ability may gain a way to express it. The result still has to earn attention.

This is where the cultural argument touches the wider moral case for AI. Human possibility should not be reserved for those who already possess every skill, every credential or enough money to hire every specialist. Assistance can let more people attempt things. Some attempts will be poor. Some will be derivative. Some may be very good.

The possibility of bad work is not a reason to forbid the attempt. We can judge the work when it arrives.

That judgment should remain demanding. An abundance of easily generated material can waste attention. A fluent paragraph can carry no thought. A striking image can rely on a worn idea. AI may make these failures easier to produce at scale. The answer is stronger taste, better selection and clearer responsibility, not an assumption that the method alone decides the merit.

Readers do not owe attention to a work because its maker used advanced technology. They also do not owe contempt.

A prize is a promise

A competition can choose to honor a particular kind of human achievement. A race can specify the equipment runners may use. A music competition can set performance conditions. A literary prize can state what forms of assistance are permitted. The rule helps define what the award means.

If a participant agrees to a clear rule and violates it, the quality of the performance does not erase that violation. A beautiful result can be ineligible.

But the existence of a rule does not make every possible rule wise. Institutions should be willing to explain what they are trying to recognize and how a restriction serves that purpose. They should apply the rule consistently, examine evidence fairly and avoid changing expectations only after a controversial entry succeeds.

There is room for different competitions. One may honor unassisted composition. Another may admit collaborative or tool-assisted work with disclosure. A third may focus on the experience created for a reader. Each should say what it is doing.

The wider culture need not adopt the eligibility rules of any one prize as a universal definition of art.

This distinction protects both tradition and experiment. People who value a particular practice can preserve a place for it. People who want to explore new methods can do so without pretending that their method is identical. The public can learn what each practice produces.

It also protects the person accused. A serious review should identify the alleged rule breach and allow a response. An accusation should not expand until it becomes a judgment on the person’s right to create. Criticism can be firm without becoming a campaign to erase someone.

In the French controversy, the disputed facts and separate allegations make restraint necessary. We can discuss the questions it raises without pretending to have resolved them. The more useful general argument rests on principles that would remain sound whether the AI allegation is eventually confirmed or rejected.

A work’s merit deserves examination. Its provenance deserves honesty. A person’s reputation deserves evidence.

A culture that can be surprised

The most revealing response to a successful new method is sometimes a refusal to be surprised.

If a result is weak, it proves the tool cannot create anything worthwhile. If a result is strong, it must be disqualified from the category. Either way, the original belief survives. The test has been arranged so that no evidence can change the answer.

That is not a defense of human creativity. It is a defense against learning.

Human creativity is strong enough to face a new instrument. It can borrow, argue, reject, transform and invent. It can also insist on fair rules for the people whose work supports the instrument. There is no reason to make cultural openness depend on surrendering those claims.

A genuinely pro-human position should be interested in whatever helps people express, understand and connect. It should ask whether the person using the tool gains power or merely rents a temporary ability from a company. It should ask whether creators can earn a living and whether audiences are being misled. Those are difficult questions with practical answers to develop.

Calling every assisted work empty answers none of them.

Suppose a poem made with assistance gives a reader language for grief. The reader may later decide that its origin changes the kind of connection they feel. That is their right. But another reader may know the origin and still find the words helpful. Their response is not false simply because it differs.

Suppose a novel is acclaimed before its method becomes controversial. If assistance is later established, we should ask what that tells us about the capacities of the tools, the judgment of the maker and the expectations of readers. We should investigate any deception or copying separately. Curiosity is not acquittal. Condemnation is not criticism.

The future of culture will be poorer if we permit only two reactions: worship the machine or burn the page.

I want better work. I want more people able to make it. I want the people who contribute to culture to have rights and a share in its value. I want readers to be told the truth when the truth matters to what they are being offered.

And when something is good, I want us to have the courage to ask how it became possible.

Sources and notes