[Journey] Creating Good Books With AI

michaelwilliams1 said:

Thanks for sharing your journey and being transparent about your process. I like that you're focusing on creating genuinely useful content instead of relying on AI-generated text alone.

One suggestion would be to validate the topic before investing too much time in writing. Looking at reader reviews, common questions, and gaps in existing books can help ensure you're addressing real audience needs. I think combining that research with your AI-assisted workflow could make the final product much stronger.

Looking forward to seeing how your experiment develops.

Click to expand...

Thanks, I appreciate the comments.

Validation definitely seems like a good idea. I’ve heard a few people mention using Amazon reviews to work out what’s working and what isn’t. Some recommend comparing the 5-star and 1-star reviews, while others say it is in the 3-star reviews because that’s where readers often explain what was missing.

Your point about looking for knowledge gaps really resonated with me as well. It like in SEO

Have you used that approach yourself?

It’s definitely something I want to explore once I get to the research stage. In fact I can image a couple of experiments that may be fun to run. At the moment I’m trying not to swim before I can paddle .

the one thing i’d add is to build an evidence pass into the workflow, like a step where you replace any vague claims with real-world examples, citations, or at least verifiable references so it doesn’t feel like generic self-help.
also, when you start picking the random topic, try to choose something narrow enough that you can fill multiple chapters with distinct subproblems

michaelwilliams1 said:

Great suggestion. An evidence pass can definitely improve credibility and make the content more useful. I also agree that choosing a narrower topic makes it easier to provide more depth and practical value throughout the book.

Click to expand...

yeah exactly, and the evidence pass is what keeps it from turning into generic “motivational” stuff. if you pick a narrow angle, each chapter can focus on a specific problem plus a real example or exercise, and that’s what makes it feel genuinely useful.

Nefertum said:

the one thing i’d add is to build an evidence pass into the workflow, like a step where you replace any vague claims with real-world examples, citations, or at least verifiable references so it doesn’t feel like generic self-help.
also, when you start picking the random topic, try to choose something narrow enough that you can fill multiple chapters with distinct subproblems

Click to expand...

Just want to check by evidence pass do you mean in the editing phase?

I agree with you both on drilling down. I really like the one person one problem approach. I think some can go too far down but as a general rule it’s a good one.

The problem is. Hollywood with their massive billion dollar budgets cant figure out how to make AI stay compliant enough to make a story thats actually engaging. Maybe you found a loophole? In that case that person needs to keep it to themselves and make some dough.

SIPSAMURAI said:

The problem is. Hollywood with their massive billion dollar budgets cant figure out how to make AI stay compliant enough to make a story thats actually engaging. Maybe you found a loophole? In that case that person needs to keep it to themselves and make some dough.

Click to expand...

I can’t image trying to write long form fiction with AI. That feels like you would be really fighting against the AI. There was a study recently that showed that AI writing of fiction can be spotted in the very structure of the story not even just the words.

I am looking at non fiction and hoping that I can create something that comes from using AI strengths.

Two of the books have arrived Atomic Habits and Brainfluence. I picked two random chapters and prompted them in Claude and ChatGPT.

One looked at the 2 minute rule and the other the paradox of choice. Both chapters were prompted the same way.

The Claude output was interesting. On the 2 minute rule it pulled in extra information and made a distinction between two different two minute rules. It handled paradox of choice much better.

ChatGPT only writes in single line paragraphs so everything looks like a massive list.

I was pleasantly surprised with the paradox chapter in Claude. It read closer to the information covered in the book.

It’s still far from perfect and no where near something that could be published. It was however a simple prompt.

I wonder why they both did one well whilst struggling with the other.I suspect it’s because the paradox of choice is more concrete in the information. Also I just noticed the paradox chapter is probably half the length that may have an impact.

AI writing is still an information dump at this point. Been an interesting start to this experiment. Will continue this compare and contrast experiment this week.

A small update after a few evenings of experimenting.

I’ve now recreated around ten random chapters from the different books, across a few different genres, using three different prompts and 2 AI’s (AIs AIses? not sure on the plural).

One pattern that’s starting to stand out is that AI seems much more comfortable with concrete, well-defined ideas than topics that are nuanced, controversial or abstract.

Another thing I’ve noticed is chapter length.

If I let the AI write as much as it wants, it has a tendency to waffle. If I force a much shorter chapter, it becomes far more focused, but often skips over important details.

As an seperate thought I also tried adding a target audience into one of the prompts. For Brainfluence I used something like “business owners wanting to improve online conversions.” Though using the one person one problem this might help

I expected that to improve the writing.

Instead it became… odd.

It reminded me of the “How do you do, fellow kids?” meme. The AI kept trying to sound like it was talking directly to business owners, but it felt forced and unnatural rather than authentic.

every experiment seems to raise another question.

I don’t feel any further forward.

I guess at the moment I’m learning at least what doesn’t work.

rexant said:

You're going to sell those on KDP or Etsy mate ?

Click to expand...

KDP will be the main place. I might look at POD later down the line.

rexant said:

I also listed many books on KDP, but no sale
Etsy is good for POD(Tshirt and digital goods)

Click to expand...

I do POD mugs on Etsy. For the book (s) I would do my own site linked to a book POD company?

Would. Love to know, how long have your books been up? How are you promoting them and did you write them or use AI?

rexant said:

No bro, same like mugs and tshirt bro, you need a supplier for printing
I've uploaded the books 2 months ago and all made by AI

Click to expand...

Sorry to hear about the books. It doesn’t seem that KDP is as easy as people make out. Did you follow any programs?

Etsy is great in the right market and niche. I mainly target woman and a sub niche within spirituality.

rexant said:

I will focus on the quantity and going to upload more, hopefully I will get some profits
You're right, Etsy is better, you making good on POD mugs bro ?
I plan to make some journeys with POD(tshirt) on Etsy

Click to expand...

Mugs are my full time income. KDP is a side interest at the moment.

I have another product I am looking to launch later this year which is POD but if it works I will transition to holding stock. Thr market up is ridiculous compared to POD on this one.

Never done the tshirt thing. Are your books and Etsy linked or completely separate

My other books arrived and I think I have found an author I really enjoy reading Erik Larson does this creative non fiction. The Devil in the White City was a really interesting read. Not sure if there’s any practical implications for writing Good books with AI.

One thing I am going to try comes from two different sources. One guy whose name I forget talked in a podcast about every page needs to earn its place in the book. I think that’s gold. I am not sure on how to implement that just yet but it feels right.

I think in someways this may tie into what Dean Koontz does with his heavy editing and reworking of every page before he starts the next. He’s a fiction writer so it doesn’t quite translate but I think that’s a way of improving a chapter.

I think I had a bit of a eureka moment today when YouTube played a random video on a phd level paper.

It got me thinking about the process of writing and so I went across to several AI chatbots to see what they would say.

The consensus was that to write a book or chapter AI has to:

1. establish the book premise,
2. understand the target audience
3. address the book purpose
4. research the subject matter
5. select what needs to be include,
6. develop a chapter skeleton
7. select a voice and tone
8. write the the chapter

The end result is that I think asking AI to write a good book or even a chapter is too much in one pass.

I am starting to wonder whether one brain for one task might be a better option. Ask chat got to research the subject first then write based on that. I guess in someways this is closer to what a human would do.

I don’t know but it’s an exciting thought and feels like my first move forward

After my last post I wanted to test whether breaking the writing process down into separate tasks helped produce noticeably better output.

For this experiment I want to do a single variable test.

The first experiment which we have done before was simply a one and done prompt Prompt: One-shot “Write a 1,500-word book chapter on saying no for a book on “people pleasing”

Then a simple structured Multi-Brain Sequence split into two prompts:

Prompt 1: Research and summarize key points, statistics, examples, and common pitfalls on the topic.

Prompt 2: Using only the research output above, write a 1,500-word book chapter on saying no for a book on “people pleasing”

So the second prompt in experiment 2 is deliberately the same as in experiment 1 but with the confined of using the research in prompt 1 of experiment 2.

I ran everything in incognito to avoid any carry-over from previous chats.

The multi-brain approach produce a much better book chapter. Notably so.

In case this was a fluke I repeated the experiment a further 11 times and consistently the multi brain method produces higher quality results.

The multi brain method versions felt like they flowed better were more specific (even directly referencing some of the research)

Now to be honest it’s far from perfect. The voice issues are still there it still reads like AI but on a side by side read there’s a clear winner.

I feel like this is a much needed win. Having listen to so many people on YouTube and social media on how easy KDP is I honestly thought I would have at least a draft of a book by now.

For now I am going to put a tick next to multi brain. Next I’m going to expand this into a longer chain and see if there comes a week point. Context window may come a problem but we will see.

Has anyone else experimented with prompt chaining like this would love to hear your experiences.

So I think I have an idea to test based on multibrain and it goes back to the books I bought earlier.

One book I liked was Brainfluence and I have kept going back to it. Perhaps to write a good book I shouldn’t be writing a traditional non fiction but rather a modular or listicle style book.

Use the multi brain technique on each chapter and end up with x amount of smaller chapters that each cover one topic.

To my mind that gets the best out of AI minimises the issue with context windows and should produce a book of substance.

I've written some long form books and Ai does best when its given a structure to follow, even then it needs heavy refining afterwards. Took me a few hours to get like 120 pages even readable to a normal person. And then a few more days in editing and changing sentences to get it somewhat okay. You also need to distill a writing style down into a prompt and give it examples to write like otherwise it just drifts into repetitive nonsense. Good luck!

Best of luck to you

Hamoudi34 said:

I've written some long form books and Ai does best when it’s given a structure to follow, even then it needs heavy refining afterwards. Took me a few hours to get like 120 pages even readable to a normal person. And then a few more days in editing and changing sentences to get it somewhat okay. You also need to distill a writing style down into a prompt and give it examples to write like otherwise it just drifts into repetitive nonsense. Good luck!

Click to expand...

I have noticed in my AI experiments so far that the flow of sentences doesn’t follow a logical or sequential pattern. For those into theology it’s a Markian sandwich. Here’s A here’s b and here’s A. Interesting and not obvious choice by the machine.

I do have writing style to complete later on but I agree it’s vital and not just putting in the style of or inspired by x. I like the idea of al lay creating an avatar for the author beyond just words they use. This is for further down the line

What did you end up doing with your book(s)?

I'm not sure AI can write a viral book because it doesn't have a unique character (yet). However, you should write the book for yourself. If you can't stop reading it, then it's fine.