I wrote and self-published a book. I used AI throughout, and I made a lot of mistakes. This is the honest account of both, because the mistakes are where the learning is.
The book is called Slow AI. It argues the thing this newsletter argues: knowing when to use AI and when to leave it alone. I used AI to help me make it. If that sounds like a contradiction, it is worth a second look, because the whole point of critical AI literacy is knowing the difference between the parts a machine can help with and the parts you have to do yourself.
In this post I will:
Show exactly how I used AI to research, shape, edit and market a book, and where I kept it out.
Walk through the whole self-publishing process, from first draft to Amazon best-seller status.
Own every misstep along the way, so you can skip them.
Why I wrote a book, and why I self-published it
Two reasons. A book is a credential, and it is an argument I wanted to make in full, beyond the constraints of a newsletter.
On the self-publishing choice I want to be straight, because I am not doing this out of principle. I have published ten books the traditional way. It is a good process and I am glad I did it. Here is the thing nobody quite tells you: even with a publisher, you end up doing most of your own marketing. So if I am going to be the one selling the book anyway, I would rather own the whole thing, move at my own speed, and keep the argument current, which is something traditional publishing is too slow to allow. A book about a technology that changes every week needs a format that can change with it.
I am still learning how to do this, and I am happy to build in public and make my mistakes where you can see them. One more thing, so there is no confusion: if this works, and Penguin rings up and asks me to publish the next one with them, I will say yes in a heartbeat. This is about learning how the whole thing works, and being honest while I do it. There are no purity tests here.
How I actually used AI
I used AI in four places, and I want to be precise, because vagueness here is how people mislead each other about what these tools do.
First, research. Chasing sources, pulling figures, finding the original paper behind a headline. Second, content strategy and publicity. I owe this one to the amazing Jen Benford, who pointed out that the tools often earn their keep around the writing more than in it. Third, editing. An AI is a reader that never tires of going over the same paragraph. Fourth, checking. I had AI verify every reference in the book against its live source, which I will come back to, because it made mistakes.
The words are mine. Every argument is mine, and every judgement. Here and there a turn of phrase was sharpened by a machine, and where it earned its place I kept it. The one thing I kept entirely for myself was the thinking. That is the whole point of the book, so it had better be the whole point of how the book was made.
The process, step by step
Here is the actual order of operations, so you can see the shape of it.
A first draft, written by me chapter by chapter, with the machine chasing sources and every factual claim checked as I went.
A review panel. Two simulated peer reviewers, a seasoned professor and a sharp early-career critic, and five reader personas, from a near non-user to a power user. I ran the whole panel twice, and I edited after each round.
A commercial test. I asked the personas five blunt questions about every draft: would you pay for this, would you be glad you did, would you share it, would you recommend it, would you write a review. I kept editing until the answer was yes to all five.
A structural fix. I moved the most positive chapter much earlier, so the book opens as a fair argument, and the harder chapters land better for it.
An economy pass. One rule: every sentence earns its keep or it goes. That cut about a hundred small pieces of throat-clearing and hedging.
Reference verification against live sources.
The extras. Poems between the chapters, written by me, a glossary, an acknowledgements page, a dedication, and a note explaining exactly how AI was used.
Formatting for Kindle. I built the book as an ePub with every source as a live hyperlink, then checked that each one actually worked.
Pricing and launch.
A marketing and publicity campaign, drafted with AI and run across every channel I have.
The prompts, for the curious
I am wary of sharing prompts. They date fast, and a good one is really just a clear description of what you want, in your own words, which is a thing you already own. So take these as a snapshot, not a recipe.
For the peer review, I used a Claude skill that runs two reviewers over the draft:
Act as two academic peer reviewers of this draft. Reviewer one is Professor Margaret Chen, a seasoned full professor: fair, rigorous, focused on argument, positioning, and contribution to the field. Reviewer two is Dr Amir Osei, a sharp early-career postdoc: meticulous and a little combative, focused on method, evidence, and internal consistency. Each writes a full structured review with a recommendation. Then synthesise the top five revisions I should make.
For the publicity, I described the job and let it work:
Act as a world-class publicist. Prepare a social media marketing campaign for my book across all my channels. Sequence it around the fixed dates I already have, set honest KPIs and challenge my targets where they are too soft, and make the branded graphics as needed.
The missteps
This is the section a reader asked me to write, so that others might learn from my experiences.
The AI wrote my own rule into the book. One of my instructions to it was to never raise a problem without offering a fix. It took me literally and wrote this into a chapter, as prose. Thankfully I caught it. The machine follows your instructions to the letter, including the ones you never meant literally.
Reordering the chapters broke things the automated checks could not see. After I moved a chapter, the book still said ‘after four chapters of harm’ at a point that was now the second chapter, and ‘the six chapters’ when there were seven. The software that checks cross-references missed all of it, because these were sentences, not links.
A statistic was wrong, and I nearly kept it. A figure about the cost of one AI product was out by an order of magnitude. Checking it against the primary source, the research paper itself, is the only reason it is not in the book.
When I checked the references I found no inventions, but nine wrong labels. Headlines paraphrased so loosely they no longer matched the article, a web address with a broken path, a URL slug mistaken for a title. AI had fabricated nothing, and yet it had mislabelled nine things. If you take one thing from this post: never trust a citation you have not opened yourself.
The AI told me it was done when it was not. More than once it made the edits I asked for but did not rebuild the file, so the version it handed back looked finished and was not. Verify the output. Do not trust the report of the output.
I forgot the page numbers. When I initially produced the paperback version of this book, it didn’t contain any page numbers. Bizarrely, some readers received a version of the book before I got my own proofs, so they were able to correct this mistake for me. The new physical version now, thankfully, has page numbers and even a table of contents.
I got the pricing wrong first time. I planned to launch at 99p using a Kindle Countdown Deal. Amazon does not allow a Countdown until a book has been on sale at a steady price for thirty days. So the launch plan changed to a simple 99p price for the first couple of weeks.
I got the reviews wrong first time too. I told my advance readers they could read the book for free and leave a review. Amazon only really counts a review that comes with a purchase, and giving away free copies in exchange for reviews breaks its rules. So I went back to them, admitted the mistake, and asked them to buy it at 99p on launch day if they were happy to, or to just keep the book as a gift, with no pressure either way.
The pricing and launch decisions
I made three decisions here:
Price. I launched at 99p, and not as a gimmick. At that price Amazon pays a 35% royalty, and I would rather the book reached people than earned more per copy. It rises to £2.99 on 28 July, which is where it will settle. At £2.99 the royalty jumps to 70%, so the book starts to earn back some of the two years of thinking behind it.
The Countdown that got away. I had wanted to open with an Amazon Countdown Deal, the official discount banner that does a lot of the marketing for you. For that reason it is now booked for late August as a second, sanctioned 99p window. A missed plan that turned into a second launch.
KDP Select. I enrolled, which is ninety days of Kindle exclusivity in exchange for the Countdown option and a few other levers. It keeps the book off other stores for now, and I decided the launch mechanics were worth it.
On Amazon, and a standing offer
Some of you will not love that this is on Amazon, and I understand why. It is a monopoly with a long charge sheet, and putting a book about thinking for yourself onto it is a fair thing to raise an eyebrow at.
Here is my honest position. I wanted to run the experiment end to end, Amazon included, to learn how the whole machine works from the inside. So far I like it, and I am not going to pretend otherwise to sound purer than I am. The part that gives me pause is the exclusivity. Kindle Select keeps the book on Amazon alone for ninety days, and handing one company that much reach is the very pattern I spend my time warning about. That is my truth. Yours may be different. Slow AI exists to help you reach your own conclusions. This is one of mine.
So, a standing offer. If you will not buy from Amazon on principle, email me at sam.illingworth@gmail.com and I will send you a version of the ePub for free. No strings, and no review expected. I would rather you read it than have a rule about where you shop keep it from you.
My book, Slow AI: Knowing When to Use AI and When to Leave It Alone, is out now on Kindle and paperback. If you read it, an honest review helps more than anything.
If you have read the book, then please do leave me a review, as this is what helps the book to travel and with it the work of independent authors like myself.
How the launch actually went
I am writing this in launch week, so here is where it stands as the post goes out.
I actually launched last Friday 10th July, quietly, a few days before I made any noise. I told some advance readers and friends over the weekend. I sent a note on the Monday, emailed my Substack list on the Tuesday, and wrote this on the Wednesday, for publication today (Friday). I updated the numbers yesterday. As things stand the book has sold 512 copies. It reached number one in Education on Amazon Kindle in the UK, number one in Human-Computer Interaction in both the UK and US, and number two in Teacher Resources in the US. It has earned me roughly £330 ($445) in royalties. I am telling you that last figure on purpose. A number one bestseller at 99p does not make you rich, and it was never supposed to. I priced it low to reach people, not to buy myself a holiday home.
Letting it go live early was deliberate. Early downloads and honest reviews do effective work with Amazon’s ranking before the main push, so by the time the email went out the book already had a foothold. Kindle and paperback are both up, my own website carries it, and the advance readers who put their hands up got it first. The campaign around it, the emails, the notes, the LinkedIn and TikTok posts, were planned in advance, which is the only reason I am not scrambling this week.
What I would tell you if you are about to do this
Use AI for the tireless work: the initial research, the peer review, the tenth read of a paragraph, the publicity. Keep the thinking for yourself. Check every fact against the source, not the coverage. Get real readers, and listen to the ones who make you wince. Own your mistakes in public, because that is where other people learn, and where you do too.
Go slow.
This is the discipline the Slow AI curriculum builds month by month: how to keep your judgement while everyone around you loses theirs. Twelve months of structured inquiry, monthly live sessions, and a community grounded in the research rather than the hype.
Congratulations on the book and the launch!
And thank you for walking through your creation process step by step. You’ve produced a wonderfully helpful guide for anyone thinking of publishing a non-fiction book on Amazon!
Congratulations, Sam!
I was following along as you were launching the book while I was in the middle of moving house. I really wish I could’ve joined in more.
I’m so glad to see it’s been such a fantastic success. Well deserved!