Kobo rejected 45% of self-published books last year, mostly over AI

Rakuten Kobo rejected 45% of titles submitted to its self-publishing service in 2025, with CEO Michael Tamblyn attributing over 80% of those rejections to AI-generated content of poor quality. The company relies on quality assessments rather than AI-detection software, which Tamblyn acknowledged is unreliable. The policy reflects Kobo's stance against the flood of machine-written manuscripts, contrasting with Amazon's disclosure-based approach.

A self-publishing platform exists to say yes. Kobo’s, last year, spent a remarkable share of its time saying no. Rakuten Kobo rejected 45% of the titles submitted to Kobo Writing Life, its self-publishing service, in 2025, and chief executive Michael Tamblyn attributes more than 80% of those rejections to books he judges to be manifestly AI-generated and of very poor quality. Hundreds of thousands of submissions, on his account, were turned away. The figures came from Tamblyn directly, first in a keynote at the CONTEC conference in Buenos Aires in April and then amplified in a post on Threads in early May. He described the platform as being on the receiving end of a firehose, a volume of machine-written manuscripts that would barely have registered as a category a couple of years ago and now dominates the reject pile. The numbers deserve care, and Tamblyn himself supplied some of it. The 45% is the share of submissions rejected, not the share of submitted books written by AI, and the 80%-plus figure applies only within that rejected set. Kobo has not published the total number of files reviewed, the full breakdown of rejection reasons, or the method it uses to decide a manuscript was machine-written. The statement reflects the platform’s posture more than it measures AI’s true share of self-publishing. That caution matters because detection is the weak link. Tamblyn has acknowledged that the AI-detection software Kobo evaluated regularly mistakes human writing for machine output, and the reverse, which is why the company has leaned on quality as the practical trigger for rejection rather than attempting to certify the precise origin of every file. A book that reads as cheap and mass-produced gets turned away whether or not its provenance can be proven. The stance is written into Kobo’s rules. A help page updated in May lists common grounds for rejection, including titles that are too short, wholly or partly AI-generated content, poor translations, and prices judged excessive. Kobo’s terms reserve the right to accept, refuse, suspend, or remove any title, state that submitted books will not be used to train generative-AI systems, and reserve the right to use AI to read and analyse texts in order to assess them. The approach is a quieter cousin of the one the rest of the industry is wrestling with. Amazon’s Kindle Direct Publishing took a different route, requiring authors since September 2023 to declare AI-generated text, images, or translations https://thenextweb.com/news/online-thieves-are-impersonating-authors-to-launder-money-on-amazon at upload, while leaving AI-assisted work undeclared. Kobo prefers the refusal to the disclosure form. Both are responses to the same pressure, the flood of cheap machine-made text that platforms once worried about in theory and now sort through in bulk, a problem adjacent to the ‘workslop’ https://thenextweb.com/news/ai-workslop-knowledge-decay-harvard-business-review-productivity spreading through the wider economy. What Kobo has not said is how many genuine books it turned away to keep the bad ones out. The cost of a market awash in generative content https://thenextweb.com/news/freepik-rebrands-as-magnific is borne partly by the writers caught in the filter. Get the TNW newsletter Get the most important tech news in your inbox each week.