We solved one of the annoying problems with AI writing: keeping context across an entire book series

Generating one book with AI isn't particularly difficult anymore.

Generating Book 2 and Book 3 without the AI forgetting what happened in Book 1 is a much more interesting problem.

We've been working on this inside Inkfluence AI recently.

The obvious approach would be to generate each book independently. But that starts falling apart pretty quickly for fiction.

A character shouldn't suddenly have a different personality in Book 2. Relationships need to carry forward. Locations and world rules can't randomly change. And something that happened in the first book needs to remain true several books later.

The problem isn't really generation. It's persistent context.

So we built a Series workflow into Inkfluence AI that starts with an existing book rather than a blank prompt.

The user can choose how the next books relate to the original:

Sequel - each book continues chronologically from the previous one.

Companion - each book explores the same topic, world or concept from another angle.

Expanded - something introduced in the original becomes a more detailed standalone volume.

The existing book effectively becomes the foundation for the series rather than asking the user to explain everything again.

That's the part I find interesting from a product perspective.

Long-form AI products have a very different problem from chat products.

In a chat, losing a small detail from 50 messages ago might not matter much.

In a novel, changing someone's history, relationship or established world rule halfway through can undermine the entire thing.

And the longer the project gets, the more context there is to preserve.

We tested the workflow by taking an existing fictional book with a protagonist called Luca and turning it into a three-volume series. Inkfluence AI created all three connected books simultaneously, after which each one could be edited, proofread and exported individually.

I've recorded the full 3-minute test here:

https://www.youtube.com/watch?v=_ffxzKXVGHk

And documented the workflow here:

https://www.inkfluenceai.com/learn/how-to-write-a-book-series-with-ai

I'm curious how other people building long-form AI products are handling this.

Once your output becomes tens of thousands of words rather than a chat response, how are you deciding what context needs to persist — and what can safely be forgotten?

Sam
Founder, Inkfluence AI