If I pick up a book, or open an eBook and see an author’s name on it, I assume that person wrote the entire book.
Sure, they may have used an editor. Hopefully someone proofread it. Maybe the publisher suggested changes. But fundamentally, I have every right to believe the words came from the person whose name is printed on the cover.
But it’s 2026, the world is going crazy over AI, and maybe that expectation just isn’t realistic anymore.
The question of whether the named author actually wrote the book isn’t hypothetical. The Wall Street Journal recently looked at the turmoil AI is causing in publishing and how difficult it can be to verify whether a book was wholly written by the person named as its author.
If I buy a book with someone’s name on the front, I want to trust that they actually wrote it. I’ve been thinking about this quite a bit because I use AI myself, and it’s creeping further and further into my creative process.
I use it when I’m putting videos together for The Weekly Spark and my other channel, The Spark!. I might ask ChatGPT to help generate an outline, then I’ll write the script and use it to check the flow. For me, AI is closer to an editor or proofreader than a writer.
I might ask it to generate a video description or suggest a few titles to A/B test based on my finished transcript. I’ll also ask it for thumbnail suggestions or use it to clean up a thumbnail I’ve created.
Ultimately, though, the thoughts and opinions are mine. I write the script and decide what comes out of my mouth. I wouldn’t be comfortable asking AI to write the whole thing and then reading it as though it were mine. If I were writing a book, I certainly wouldn’t want AI generating an entire chapter and then claiming I’d written it.
Amazon, believe it or not, makes almost exactly this distinction.
Under Amazon’s Kindle Direct Publishing content guidelines, AI use is separated into two categories: AI-generated and AI-assisted. If a tool creates the actual text, Amazon calls it AI-generated, even after substantial editing. If you write the text and use AI to refine it, check it for errors or brainstorm ideas, that’s AI-assisted.
Amazon doesn’t require authors to disclose AI-assisted use. It does require publishers to tell Amazon about AI-generated content, but that doesn’t mean shoppers see an AI label on the book’s product page. The buyer may still have no idea who wrote the words.
So how do you prove who wrote something?
AI-detection software might sound like the answer, but it isn’t all it’s cracked up to be.
Back in 2023, Ars Technica tested part of the United States Constitution using GPTZero and ZeroGPT. GPTZero said it was likely written entirely by AI. ZeroGPT also classified parts of it as AI-generated.
Now, unless the people who wrote the US Constitution had access to ChatGPT about 235 years before the rest of us, we’ve got a bit of a problem.
They tried passages from the Bible too. A passage from Genesis was flagged by ZeroGPT as 88.2% AI-generated. Yet I think we can be reasonably confident that Moses wasn’t sitting there with Claude open in another tab.
There’s a more recent example too. In May this year, the Authors Guild, America’s oldest and largest professional organisation for published writers, tested five detectors using ten articles published before generative AI was widely available.
One detector called all ten mostly AI-written, including two at 100%. Intriguingly, other detectors called the very same articles entirely human-written.
That doesn’t mean every AI detector is useless then, but a score clearly shouldn’t be treated as conclusive proof. If AI can’t reliably identify AI-written text, you haven’t got much hope of spotting it because an author used an em dash or, shock horror—the word “delve”.
An em dash is punctuation, not a confession, and “delve” is still in the dictionary.
This concerns me because I could write something myself, use AI to proofread it, accept a few good suggestions and then have a detector decide AI wrote all of it.
Another interesting development is Anthropic’s plan to add invisible watermarks to text produced by future Claude models. Anthropic says the watermark can indicate that Claude was likely involved with the text, but it cannot distinguish between “Claude wrote this” and “Claude heavily edited this”. Nor does it prove anything about ownership or authorship.
To be fair, Anthropic also says that if Claude only corrects a handful of spelling, grammar or punctuation mistakes, there may not be enough of a watermark to detect at all.
But what happens when the editing is more extensive?
If I give Claude something I’ve written and ask it to improve the flow or remove repeated words, some of Claude’s choices could carry that watermark. If somebody then treats its presence as proof that AI wrote the whole thing, despite Anthropic explicitly warning that it proves no such thing, how is that fair?
The Authors Guild has tried to help by introducing Human Authored certification: a badge saying that a human wrote the book. The text must be human-written, although the scheme allows a very small amount of AI modification through tools such as spelling and grammar checkers.
But where exactly is the line?
If I write 80,000 words and AI spots my spelling mistakes, surely I wrote the book.
But what if I give AI the plot and ask it to write a chapter? What if I then rewrite half of that chapter? What if I rewrite 90%? At what point does it become mine?
I don’t know. I’m not sure anybody has a particularly good answer yet.
This may surprise you, but I don’t necessarily object to an AI-generated book. If it were clearly labelled, genuinely excellent and something I enjoyed reading, I’d have no problem with that.
What I would object to is being misled. If I bought a novel because someone’s name was on the front, then discovered they’d asked AI to write it, I’d feel cheated.
The type of book matters too. With a reference book, I mainly need accurate facts and information, so I’m less bothered about having a connection with whoever constructed the sentences.
But a memoir? That’s completely different.
If I’m reading someone’s autobiography, surely part of what I’m buying is access to that person’s recollections, thoughts and voice. Imagine discovering halfway through a memoir that most of those deeply personal passages had actually been generated by AI.
At that point, what exactly am I reading?
Maybe that Authors Guild badge isn’t such a silly idea after all.
I’d genuinely like to know where you draw the line. If somebody comes up with a book’s characters, story and plot, but AI writes the words, would you still call that person the author?
Let me know what you think in the comments.
Just don’t shoot the messenger.
If you’d like to support my writing, you can always buy me a digital coffee. It genuinely does help keep this little corner of the internet ticking along:
Connect with me: X | Instagram | Threads
Subscribe to my YouTube channels: The Spark | The Weekly Spark
Buy stuff: Affiliate Page
