When John Degen’s murder-mystery novel Seldom Seen Road is published next month, he’ll be one of the first Canadian authors to have a small label with the words “Human Authored” printed on the back of the book jacket.

The certification — developed by the Society of Authors, a U.K. trade union — operates on an honour code system: If you’re an author who wants to declare that your book was written without any assistance from artificial intelligence, this is one way of doing so.

“I really wish it wasn’t necessary, I’ll say that,” said Degen, a Toronto-based writer who is president of the Writers’ Union of Canada and chair of the International Authors Forum. “But because it is necessary, I’m very proud to stand behind my work.”

The label is nothing if not timely. Weeks ago, the North American publishing industry was rocked when a New York Times story accused horror writer Mia Ballard of using generative artificial intelligence to write her novel Shy Girl.

The Times presented evidence compelling enough for Ballard’s publisher, Hachette, to cancel the book’s U.S. and U.K. release entirely. Ballard denied using AI to write the novel, but said it was possible an editor she’d worked with on the self-published version might have.

The scandal divided the industry. Some accepted the accusation — and Hachette’s response — as truth. Others felt the punishment was prejudicial, since AI-detection software like the kind used to evaluate Ballard’s writing tends to be imperfect.

The incident, which Ballard says has ruined her career, demonstrates the conundrum that literary professionals face as they comb through every pitch, query letter and manuscript lobbed their way: how do you separate the proverbial wheat from the AI-generated chaff, and what happens if you get it wrong?

“I really would have preferred to see her publisher stand up for her and stand behind the work themselves because they trusted their own process,” said Degen. “I mean, the best AI detector in the world — the best detector of bad writing — is a good editorial process.”

Is there a right way to use AI?

Most of the literary professionals who spoke to CBC News for this story said it matters how AI is used in the writing process. It’s a distinction that several self-publishing platforms have also made.

For example, some distinguished between a text that is completely AI-generated (for example, when text or content are created entirely from scratch based on a prompt) — versus a text that is AI-assisted (when a human is responsible for the output, but AI is used for tasks like spell check or editorial feedback).

Kindle Direct Publishing, for example, asks writers in its content guidelines to declare any AI-generated text, photos or translations when publishing or re-publishing a book. AI-assisted works get a pass.

IngramSpark, another self-publishing tool, says on its website that it will remove any content generated by AI or “mass-produced processes,” but AI-assisted works aren’t mentioned.

“I don’t necessarily see it as AI is good, AI is bad,” said Chandler Supple, the San Francisco-based chief technology officer of River AI, a platform where writers can use AI to edit and proofread large-scale projects like books.

“People on our platform spend literally hundreds of hours working on these books. They’re not just saying, ‘Hey, Claude, can you write [200 pages] for me?'” he said, referencing tech firm Anthropic’s generative AI platform.

River AI was founded last May, and of the 20,000 people who currently use the service, about 20 per cent are Canadian, according to Supple. Many of them are self-publishing on the platforms mentioned above.

“I think that AI is a tool that people can apply to their craft to produce better results, higher quality, faster rate of output. And if you’re doing it right, I don’t think that it degrades the creative nature of the work that you produce,” he explained.

Publishers, agents fielding AI slop

Still, there are varying degrees of comfort over the use of AI in published writing, whether generative or assisted. And the risk of a potential copyright violation — an issue currently at the centre of many a lawsuit — weighs on the agents tasked with selling their client’s books.

“I typically tell people just stay away,” said Cecilia Lyra, a Toronto-based literary agent with Wendy Sherman Associates and co-host of popular podcast The Sh*t No One Tells You About Writing.

Lyra said she’s recently noticed an uptick in “undercooked” submissions from writers — usually those who are not career-professional authors — and she suspects that some of them are AI-generated. She still reads at least a little bit of every submission she receives.

“It’s my job to filter it. That burden is on me,” said Lyra. “It has become more time-consuming, but I honestly think that that is just something that, as an agent, I have to accept.”

A similar surge of AI-generated writing has put Canadian e-book company Kobo “on the receiving end of a firehose,” said CEO Michael Tamblyn during a recent interview.

In addition to selling e-readers, Kobo holds millions of e-book titles in an online library and hosts a self-publishing platform for authors called Kobo Writing Life. It’s through this channel that the company is noticing a stark shift.

The company rejected nearly 45 per cent of the books submitted to its self-publishing program in 2025. About 80 per cent were rejected because Kobo suspected they were largely or entirely AI-generated, which would have “barely been a factor” in previous years, according to Tamblyn.

“We are receiving increasing amounts of content that, as far as we can see, is likely being either largely AI-generated, partially AI-generated or entirely AI-generated,” explained Tamblyn. “And that’s coming in through the same conduit that regular authors are using to get their books out.”

Bowker, the leading information agency for the U.S. books industry, recently released data which showed that number of self-published ISBNs for fiction titles rose sharply between 2024 and 2025 — from 306,781 to 477,104 books.

A spokesperson told CBC News it might be reasonable to assume, but not prove, that AI tools made it easier for creators to publish their stories.

Kobo is focused on rooting out the most egregiously AI-generated works from its self-publication platform (like a book that aggregated 10,000 apple pie recipes from across the internet). But it hasn’t been easy to develop a broader set of guardrails because the technology can be applied in different ways, Tamblyn acknowledged.

“Do people want to know whether the books are human written or not? And how do you go about that process of flagging or asking authors to certify or asking publishers to certifying?” he said. “That is far from being a settled issue right now.”

Lyra, for her part, says she skeptical of “AI optimists” who think the technology can generate the same kind of work a human can: “How? Explain to me how. Explain to me how something that doesn’t feel can write great story.”

“I don’t get it, I don’t believe it. And I don’t know if I’m right or not, but I know it’s how I feel.”