On this week's Self-Publishing News podcast, a hoarse ALLi News Editor Dan Holloway covers Amazon's new KDP limit of two titles per week per format — down from ten, and another move aimed at the flood of AI-generated content. He also looks at a study suggesting AI detector Pangram may be more accurate than its critics claim, and at the SCBWI's tightened AI policy, which will ask award entrants to document the human creative process behind their work.
Listen to the Podcast: Amazon Cuts KDP Uploads to Two Titles a Week
About the Host
Dan Holloway is a novelist, poet, and spoken word artist. He is the MC of the performance arts show The New Libertines, He competed at the National Poetry Slam final at the Royal Albert Hall. His latest collection, The Transparency of Sutures, is available on Kindle.
Thoughts or further questions on this post or any self-publishing issue?
If you’re an ALLi member, head over to the SelfPubConnect forum for support from our experienced community of indie authors, advisors, and our own ALLi team. Simply create an account (if you haven’t already) to request to join the forum and get going.
Non-members looking for more information can search our extensive archive of blog posts and podcast episodes packed with tips and advice at ALLi's Self-Publishing Advice Center.
Read the Transcript
Dan Holloway: Hello, and welcome to another week's Self-Publishing News. And as you can probably hear, I am in the process of losing my voice as the new term has started. Students are returning, bringing all kinds of delicious viruses with them. And I will endeavor, therefore, to reach the crucial bits of the news before I succumb.
Amazon Cuts KDP Uploads to Two Titles a Week
But apologies that this will be shorter than usual. But the main news comes from Amazon, and it is news from KDP that the number of titles you are able to upload through KDP has been limited, as of Monday, September the 21st, to two new titles per week per format. So this is a significant change.
So the last limit came into effect last year, and that was a 10-title limit. And so obviously this is now a significant lowering of that. So when it says per format per week, this means if you have two titles to publish, you can do so in e-book, paperback, and hardback. But then you have to wait until midnight the following Sunday for your allowance to tick back over and another two titles to be submitted.
So obviously this has raised some eyebrows, in particular for people with reverted rights, for example, or large backlists that they want to start publishing. The algorithm obviously doesn't care, as we already know, what your reason is. It just cares how many titles you are uploading and when you are doing so.
So it's not a case that you can plead your way out of it. It is simply a case that you need to wait each week until the next week to carry on uploading. So the rationale for this is probably not an unexpected one. And it is the rationale, in Amazon's words, that it is to help maintain a positive publishing experience for all authors.
That is very similar to what Michael Tamblyn said about Rakuten Kobo's decision to turn down 40 percent of the titles that people were attempting to upload there, which he put down to the proliferation of AI-generated content. And he, as you will remember, made a point that he had taken a decision to adopt a writer-first approach.
So he pointed out that they have limited resources. They want to dedicate those resources to human authors, and the only way of guaranteeing they can do that is to turn down a significant number of titles that are submitted. So this appears to be a similar sort of rationale. We can only offer support to authors if we limit the number of titles that can be uploaded.
So obviously this comes in the wake of last year's news that a million more titles were published in 2025 than in 2024, and this was put down to AI. So, it's a very small announcement. There isn't much on it on the Amazon page. But it is a very important announcement for many people who quite possibly are going to be wanting to upload more titles than that.
Study Tests How Accurate AI Detectors Really Are
So that sort of brings us onto the inevitable AI discussion and discussion of AI-generated content and a very interesting study which I came across. Thank you to Jane Friedman for this. And this is a study that looks into how accurate these AI detectors that have been going around and have been the cause of many of the allegations of AI use.
So how accurate are they? So most articles that talk about whether an author has used AI just because they have alleged to use AI come with huge amounts of caveats, quite rightly so, about the fact that AI detection software isn't fully accurate. But this is a study into actually how accurate it is.
So this comes in an article that was published by Vauhini Vara, and it looks at, amongst other things, Pangram. So Pangram is the main AI detection software that people are using. It contains some interviews with senior people within Pangram, who are obviously going to talk about how accurate their software is.
But what it really doubles down on is false positives and how many false positives have been generated by AI software, and it does so in a really interesting way. So the author of this paper asked Tuan Chakrabarti, who you will remember is the author of the really controversial study looking at the number of titles that showed evidence of AI use.
And what Chakrabarti did was to test 50 self-published titles that came out before the release of ChatGPT. Because obviously, if they were generated, the theory goes, before ChatGPT was released, then they're not going to be AI-generated. So therefore, if any titles are flagged as AI-generated, that suggests that Pangram is giving some false positives.
And what was found was that none of these titles came back with positives. So that is a very interesting indicator, an initial indication that the claims of Pangram's insiders might have something to them in suggesting that it's actually more accurate than the stories would have you believe.
So obviously, it's not conclusive. It's one study. Caveats always remain. Not everyone uses Pangram, and as I say, it's only one study that is saying this. But that study does suggest that AI detection software is more accurate than a lot of people have been saying. So interesting stuff.
SCBWI Tightens Its AI Policy for Members and Awards
And finally, also on the subject of AI, the Society of Children's Book Writers and Illustrators has added a chunk of new policy on AI to their existing AI policy. So there are six new things that authors and illustrators must comply with, and they are designed to cut down on AI use. And so members must tick a box saying they have not used AI in any way, shape, or form.
And the society's awards — this is a really interesting one — are going to be asking award entrants to keep records to demonstrate the humanness of the creative process. So this is something we've seen already with the Commonwealth Book Prize, that one of the ways they got over the controversy around titles that had been submitted to and had won awards was to check for author's notes and the kind of stuff that you would produce in the process of writing a piece of creative work.
And so, the Society of Children's Book Writers and Illustrators, for their awards, are going to be doing the same thing. They're going to be looking at basically the notes you make along the way to establish that the authors and illustrators who enter and win their awards are actually human authors and illustrators.
So with that, I will leave you with what voice I have and very much look forward to speaking to you in more forceful tones, I hope, at the same time next week. Thank you as always, and take care.
