The Words Didn’t Change, Regardless of Who Wrote Them

A novel earns glowing reviews. Readers buy it. A prestigious literary prize puts it on its shortlist. Then someone runs the text through an AI detector, and suddenly the book that people were celebrating becomes a book they are supposed to question.

The sentences have not changed. Neither have the characters, the story or whatever moved those readers in the first place. What changed was what people believed about who, or what, helped write it.

According to the New York Times, that is what happened to Haitian-Canadian author Thélyson Orélien, whose novel, It Was Either That Or Die, was removed from the Goncourt shortlist following allegations that artificial intelligence played a substantial role in its creation.

Orélien denies using AI, and his French publisher says he completed a draft in 2019. I don’t know whether he used AI in the published novel. For the argument I want to make here, let’s suppose he did.

Then answer this: If you thought the writing was good before learning AI helped, explain what made the writing worse.

I suspect that is a harder question than running another detection test.

What Exactly Are We Judging?

When someone recommends a novel to me, I want to know whether the story is compelling, whether the characters feel real and whether the author has something to say. When I read an opinion piece, I want to know whether the argument holds up. Are the facts right? Has the writer considered the obvious objections? Will I come away understanding something better?

A piece can fail every one of those tests without a machine ever touching it. It can also pass them with assistance.

Of course readers can care about an author’s background and creative process. Lived experience can give a work particular meaning. There is nothing unreasonable about wanting to understand where a story came from. But learning that an author received help does not, by itself, tell us whether the story succeeds.

If a passage moved you on Tuesday, the discovery on Wednesday that AI helped edit it does not reach backward and cancel your response. You can reassess the author’s honesty if something was concealed. You can ask what the assistance involved. But you still have to explain what is wrong with the passage.

Too much of this debate skips that step. “AI was involved” becomes the verdict, and everyone is excused from doing the harder work of criticism.

I have written about this before, including in We Were Promised Flying Cars. We Got AI Writing Detectors. Now, having gone through the publishing process for my own forthcoming book, The Indispensability Trap, I have my own experience to add.

My Name is On the Cover

I used AI while working on my book. I also worked with a professional human editor and had several humans, in addition to me, make substantial edits to the manuscript. The ideas, thoughts, arguments and direction are mine. Most of the words are mine.

ChatGPT, Claude and Perplexity helped with research, suggestions and sharpening ideas. I thank them in the acknowledgments. I saw no reason to hide their involvement, and I see no reason to apologize for it.

The publishing platforms I encountered asked plainly whether AI had been used and about its level of involvement. I answered truthfully. I even asked Claude how it would describe the role AI had played. This was its response:

“AI tools were used to assist with copyediting, fact verification, and cover image generation. The manuscript was written and edited by the author with a professional human editor.”

I asked because I wanted to describe the help accurately. Claude’s answer did not relieve me of checking the manuscript or deciding what to disclose. If a fact in the book is wrong, telling a reader that AI checked it would be a pretty poor excuse.

In the acknowledgments, I also thank the tools for helping with research, suggestions and sharpening my thinking. I want readers to know how I worked. Then I want them to decide whether I gave them a book worth reading.

Read The Indispensability Trap and tell me where my argument breaks down. Tell me what evidence I overlooked, what conclusion I pushed too far or what I failed to explain. Those are discussions I am prepared to have. They would tell me something about the book and might help readers decide whether it deserves their time.

Pointing to my acknowledgment of AI assistance accomplishes none of that.

I put my name on the cover because I mean what the book says. I am responsible for the choices that made it into the final manuscript. I cannot claim its insights as mine and then hand its mistakes back to Claude.

In The Pangloss of Pangram, Reid Hoffman argues that detection does not establish whether writing is worth reading. He also insists that the person publishing it must be intellectually present in the work. As he puts it: “Regardless of the tools used, a human being must mean what is said.”

I agree. Publishing material you have not bothered to read or weigh fails that standard. So does putting your name on an argument you do not understand and cannot defend. Using a tool does not give you a pass on knowing what you are publishing.

Brad Feld addressed this in The Argument Against AI Writing Is At Least 2,400 Years Old. He traces the fear of weakened thinking back to the objection to writing itself recounted in Plato’s Phaedrus. Feld recognizes that writing helps people think, while rejecting the conclusion that AI must therefore never participate.

Writers should ask whether a tool improves their thinking or allows them to avoid it. The answer depends on how they work. A blanket prohibition cannot make that judgment for them.

Honesty Without the Scarlet Letter

I believe authors should disclose AI involvement. If you conceal it, or falsely claim you did everything without assistance, you have created an issue of trust. When a publisher asks what you used and how you used it, answer honestly.

But we need to allow that honest answer without treating it as an admission that the work is worthless.

Research assistance, brainstorming, copyediting and generating substantial portions of a manuscript are different activities. Readers are better served by a clear explanation than a label that lumps them together. Human editing and AI generation are not interchangeable, either. Explaining the actual process gives people something meaningful to assess.

The plagiarism allegations are another matter altogether, and I don’t include them in this discussion.

We could spend the rest of this article arguing about how accurate AI detectors are. It would not answer the question I am asking. Even a perfectly accurate detector could not tell us whether a character is convincing or an argument is sound. We would still have to read the book and make up our own minds.

If every acknowledgment of assistance brings an automatic accusation of fraud, we should also consider what behavior we are encouraging. Honest writers make themselves targets while less honest ones have a reason to stay quiet. I would rather have authors tell us how they worked than teach them that admitting they used a tool is the quickest way to get their work dismissed.

Literary societies and prize committees have every right to establish their own eligibility rules. If an organization wants to recognize only writing produced without AI, that is its prerogative. Make the requirements clear in advance, explain what counts as assistance and apply them fairly.

But eligibility for a particular prize cannot settle the quality of a book for everyone.

The Times describes one publisher’s approach that would exclude AI assistance even with brainstorming and structure. An organization can choose that boundary. Readers can also ask whether it measures what they value in literature.

Some human-only awards may endure and attract devoted audiences. Good for them. Others may eventually become quaint reminders of a world of horseless carriages, gas lamps and telegraphs, preserving a distinction that matters less and less to the people actually reading books.

I respect an author who chooses to work without AI. That choice does not automatically make the resulting prose better than someone else’s. The work still has to earn its audience.

We should expect writers to tell the truth about their process, take ownership of their ideas and answer for their words. We should also expect critics to explain their judgments. Calling something AI-assisted is a description. Making the case that it is shallow, inaccurate, derivative or badly written requires reading it.

I want my book judged that way. I would extend the same consideration to another author, including one whose methods differ substantially from mine.

Go back to the passage you admired before someone showed you a detector score. Read it again. Tell me whether the character suddenly became less believable, the observation less perceptive or the language less moving. If you found a flaw you missed the first time, explain it. That is a discussion worth having.

If you thought the writing was good before learning AI helped, explain what made the writing worse.