The Frankenstein test

I recently came across a Tumblr post showing a paragraph from Mary Shelley’s Frankenstein run through ZeroGPT, an AI checker, with the result coming back as 100% AI-generated.

Curious, I decided to replicate the experiment. I submitted the same paragraph to ZeroGPT and got the same result. I then ran it through Grammarly’s AI checker, which is often presented as more accurate, and it still came back as 48% AI-generated.

So apparently Mary Shelley writes like ChatGPT.

The real problem isn’t accuracy

That would be funny if it were not pointing to something uglier.

The problem is no longer just that AI detectors are unreliable. It is that they are training people to become suspicious of writing itself.

Because once a nineteenth-century novel can be flagged as artificial, the issue is no longer detection.

It’s distrust.

What AI detectors actually do

AI detectors do not actually detect authorship.

They estimate the likelihood that a piece of writing resembles AI-generated text.

Grammarly says this outright:

no AI detector is 100% accurate

detectors do not track authorship

human-written text can be flagged as AI-generated

lightly edited AI text may evade detection

In other words, these tools do not prove who wrote something.

They produce a score — and people mistake that score for evidence.

When writing itself becomes suspicious

And that matters because writing is not a fingerprint.

A clean sentence is not a confession. A formal tone is not proof of machine involvement. A polished paragraph is not a chatbot in a trench coat.

If anything, the Frankenstein test shows how quickly “this sounds artificial” becomes “this sounds too structured, too literary, too controlled.”

Good writing starts to look suspicious simply because a machine can imitate some of its patterns.

This isn’t theoretical anymore

That problem gets even messier once we move from detectors to actual publishing.

AI use in writing is already real enough that major publishers and academic presses have formal disclosure policies for it.

Elsevier allows generative AI tools with human oversight and disclosure. Oxford University Press requires authors to declare AI use and emphasizes transparency, quality, and trust.

So this is not some far-off scenario.

It is already here.

Authors are already using AI

Rie Kudan, winner of Japan’s Akutagawa Prize, said part of her novel was written with help from ChatGPT, with roughly 5% of the final text taken directly from the model.

That is not rumour. That is an author saying it plainly.

There have also been cases where authors appeared to leave AI revision notes or prompt-like text inside published books.

Readers found such material in works by Lena McDonald and K.C. Crowne. In McDonald’s case, a revision note referencing another author’s style was left in the text. She later acknowledged using AI to help “edit and shape” parts of the book because she could not afford a professional editor.

When suspicion has consequences

And then there is what happens when suspicion goes public.

In March 2026, Hachette pulled Mia Ballard’s forthcoming horror novel Shy Girl after a review found evidence of AI involvement.

Ballard denied using AI to write the book and said an editor had used it on an earlier version.

Whether one sees that case as proof, misunderstanding, or cautionary tale, it shows something important:

Accusations of AI use now have real consequences in publishing.

The gap between proof and suspicion

This is where the real danger begins.

If detectors cannot prove authorship, and readers cannot reliably tell whether a text was human-written, AI-assisted, or machine-generated, then something else fills the gap:

Suspicion.

And suspicion is a terrible editor.

It flattens everything:

brainstorming

editing

outlining

drafting

into one accusation.

The literary witch hunt

In publishing, this creates the conditions for a literary witch hunt.

Because once the question becomes “Can you prove this was written by a human?”

the burden shifts in a way no writer can satisfy.

If the prose is awkward → suspicious. If the prose is too polished → suspicious. If you disclose AI use → condemned. If you don’t → accused of hiding it.

Honesty becomes risky. Silence becomes suspicious.

And no detector can fix that.

The ground is already shifting

The Authors Guild has made it clear that writers already see AI as a major professional issue.

Even Amazon’s KDP disclosure policy has been described as only a first step.

That alone tells you something fundamental:

The ground is moving under authors’ feet.

The uncomfortable truth

The problem is not just that some authors use AI.

It is that the only time readers can be sure is when someone forgets to delete the evidence.

Final thought

In trying to prove what is human, we may end up making human writing itself suspect.

Sources / further reading