(Please note: this post was written and scheduled before the big blowup about the changes to Perplexity’s terms of service, and the automation work described here entirely predates the implementation of those TOS changes on 1/23/2026. I can no longer suggest Perplexity as a support tool for this kind of work, but am mentioning my use of it for transparency’s sake.)

After making  Claude.ai my virtual secretary, genre cheerleader, and typo spotter, the next logical step was automating the repetitive bits. The main reason to go to automations in the first place for certain forms of work is that the AI chatbots cannot hold a 50K manuscript (the length I mostly write to) in its memory. I conducted a few early experiments on Make.com. One of these automations analyzed public domain mysteries from the Golden Age to get a feel for the plot structure. Another was designed to give Amazon genre and SEO advice for my own books. Others helped me pinpoint quotes from the Hunter Healer King trilogy to share on social media, and scenes from the books which might lend themselves to book videos.  

By January 2026, I was burning through Make.com’s free account limits far too fast. Through the Nerdy Novelist on Youtube, I heard about n8n, an open-source automation tool which can be self-hosted on the user’s computer. The only cost for running the automations would be the API credits  spent on the ai models of my choice at OpenRouter.ai. 

Setting up n8n on my elderly computer was an odyssey in itself. I prompted Perplexity.ai, an ai wrapper tool noted for its ability to search and synthesize web results, to walk me through the process and help me troubleshoot. (Unfortunately, in the weeks after I worked with Perplexity on this, Perplexity’s owners pushed an extremely restrictive change to their terms of service and I can no longer recommend this particular ai tool). Somehow, my computer and I survived, and emerged with n8n running. 

My prototype was based on a workflow demonstrated by the Nerdy Novelist, which generated what he called a Summary Plus. It has the automation go over a manuscript chapter by chapter, summarizing each scene, cataloging the characters in each scene and their priorities, describing any setting/worldbuilding details that come up, and noting the three top tropes and three top quotes or passages from each scene. Then have it write to a separate document. The YouTuber had designed this as a quick way of extracting information from a manuscript for reference purposes, either in writing sequels or preparing a book’s marketing campaign. I didn’t need it for that purpose just then, but I could see other uses for the basic architecture. 

Building the Summarizer 

I’m not going to get into this in depth, except to say that if you’re interested, poke around this playlist on youtube to see if n8n is something you can get your head around, and whether any of the use cases work for you, and then go to the chatbot of your choice and get it to walk you through building what you want.

Expanding Horizons 

With Perplexity’s help, I found all kinds of uses for this basic “go over chapters to execute instructions” structure.  

Chapter Scoring outputs to a google spreadsheet. It grades each chapter from 1 to 10 on Plot Intensity, Character Intensity, and Ease of Reading, along with a one-sentence explanation. Download, insert a line graph with the values, and you have an instant map of the story’s plot and emotional pacing. It needs a pretty high-end AI model to do anything useful, but because the output is so minimal, it’s a fairly budget-conscious way of using those more advanced models. 

Manuscript Multi-tool lets you pick what the automation will do via form: line edits, book video ideas, social media quotes (complete with hashtags!), and more. 

Summarizer with Continuity reads what is already written in the output document and checks what it has seen in the latest chapter against that…so long as the manuscript is less than 65K-70K.  

Json Story Bible is a version of the Summarizer with Continuity. It outputs book information in json form, which you can feed to a chatbot when you are working on a sequel/prequel to the source book. 

DevEditor (really more of an alpha reader) is also built on the Summarizer with Continuity. (Probably could have been done with the baseline summarizer architecture instead). It offers options for reviewing by scene structure, pacing, character development, etc. It’s not remotely human-level, but I’m uncomfortable inflicting my work on other humans at such an early drafting stage. And, also, I never heard of a developmental editor who would scour your manuscript looking for things you needed to fix and offer how-to suggestions for $1.37 a pass. 

Most of these builds resulted from me brainstorming possible use cases with Perplexity, and then have Perplexity write the code node content, followed by me assembling the nodes, testing the results, and troubleshooting with Perplexity. I can post some code node examples if there is interest, but really, if you’re good at getting what you want out of a chatbot, you can probably do just as well on your own with Grok, Gemini or Claude. I listed out what I’ve built more to show you what is possible. And as always, don’t put blind faith in any AI’s output. Look at what it’s telling you, and do a gut check on whether that’s valid or not. 

Incidentally I just ran DevEditor in early February with my newly completed space regency, and now I’ve finished a revision pass on that manuscript. Normally, I would still be laboriously rereading the manuscript for issues. Let’s see where this goes!