AI Authored Books: The Witch-Hunt

The AI Detection Problem Nobody Wants to Talk About

As the debate over AI use in writing has intensified, detection tools have emerged as the proposed solution — the way to separate the "real" authors from the cheaters, the human from the machine. Publishers are eyeing them. Writing contests are considering them.

Online communities are using them to call out suspected AI users with the confidence of a verdict. Almost every author I know has now been accused of using AI to write their books at least once. (Even me.... a reviewer accused me of having an AI-written book that... get this... was published in 2019. Seriously?)

There's just one problem. They don't work. And not in a "needs improvement" kind of way — in a fundamental, built-into-the-premise kind of way.

The Authors Guild put it plainly in a recent article: Can AI Detectors Be Trusted? (May 26, 2026):

"AI-detection tools are AI models trained to recognize statistical patterns associated with large language model output, such as sentence rhythm, vocabulary distribution, and predictability of word choice. But polished, edited prose written by experienced human writers shares many of those same characteristics, because LLMs were trained on polished, edited prose written by experienced human writers. The more refined and controlled a writer's style, the more it may resemble the output these tools are designed to flag."

Read that again, because it's worth sitting with.

The tools were trained to detect patterns found in LLM output. LLMs were trained on polished human writing. Therefore, the tools are — by design, not by accident — most likely to flag polished human writing.

In other words, AI detection tools don't detect AI. They detect good writing.

And that quietly undermines the entire premise of AI detection as a solution to anything.

Think about what that means in practice. The experienced professional with a refined, consistent voice — the author who has spent years developing control over sentence rhythm and vocabulary — is more likely to trigger a false positive than a sloppy writer producing inconsistent, poorly edited prose. The very qualities that define craft become the evidence of suspected fraud.

The witch hunt isn't just flawed. It's backwards.

This matters because real consequences are flowing from these tools. Manuscripts are being flagged. Authors are being accused. Submissions are being rejected. And all of it is happening on the basis of statistical pattern-matching that cannot reliably distinguish between a twenty-year veteran novelist and a chatbot — and may actually be better at flagging the veteran.

There's no standard for what a "positive" detection result means legally or professionally. There's no appeals process. There's no acknowledgment, in most cases, that the tool has a known and fundamental flaw baked into its very architecture.

If you're a writer with a polished, consistent style, you are not safe from false accusation simply because you've never touched an AI tool. And if you're using these tools to check whether authors are 'cheating' — before you post that accusation publicly, you owe it to that author to understand what the Authors Guild's own testing revealed: not all tools are equally unreliable, but even the better ones are built on a flawed premise — and the worst ones are so inaccurate they flagged articles written years before generative AI existed.

The conversation about AI in writing is complicated, evolving, and worth having honestly. But it cannot be had honestly if the enforcement mechanism being reached for is broken at its foundation.

We deserve better than a witch hunt that penalizes craft.