I used a different prompting framework for this fanfic than I did for Sense and Sensibility and Placage. Around the time I started work on this, the Future Fiction Academy open-sourced a set of AI constraints they called the Narrative Physics Engine. (At this link, search for “worksheet” and then for “section_18”. Those are the important parts.) If you’re looking at the stuff at the link and thinking “gee all that seems like overkill just to scratch a fanfic itch,” that’s also more or less what Claude said when I copy-pasted all that stuff in.

We worked together on this set of constraints to build three sets of markdowns. One to use when brainstorming original fiction, to build a source document Claude could refer back to when I fed it bits of WIP for dictation cleanup, etc. One to use at the start of the revision process, to help Claude understand the finished first draft. A third and shorter one for fanfic, which I used for this fanfic. Claude built a usable outline for The Rector’s Other Business in short order, and drafted the whole thing across a handful of sessions. I was not wild about the style Claude used for this, even with an initial discussion about Collins’s voice, and a reference to the film which had initially given me this story idea (Night Creatures, Hammer’s version of Dr. Syn with the serial numbers filed off). If you’ve ever read parodies of Rudyard Kipling by the generation or two of authors immediately after him, Claude sounded kind of like that in this story, with a lot of pseudo-profound, pseudo-snarky tautologies. Because this was a shorter and lighter fanfic than Sense and Sensibility and Placage, I edited it much more heavily at a sentence level. You will note that Claude made very free with the em-dashes for Mr. Collins’s rambling speeches “in character” as the pompous clergyman he is pretending to be. I left them alone because this was a situation where they actually worked and were even period appropriate.

In terms of other stuff that are me rather than the AI: the choreography of Elizabeth’s rescue is the big one. I worked hard at prompting Claude through several drafts of that scene, and then ended up reworking it again in the edit. It was originally set at a farmhouse controlled by Annesley’s men, but I decided it made more sense logistically for Collins, Darcy et al. to overtake the ruffians on the road. Collins’s sword-cane was something I originally prompted Claude for in an early draft of that scene, and then added mentions of it in other parts of the story. Anne DeBourgh the authoress was also my idea. Claude had described her as seeming mentally absent, and then had Charlotte draw her out and discover that Anne knew at least something of what was going on. This was a bridge too far for me, in terms of People Who Know About The Smuggling. So I made her a writer instead.

The romances were the other thing I took an active hand in. Claude did a decent job on the Collinses overhearing Darcy’s proposal at Hunsford, and Darcy’s emotional state immediately after that. Most of the details of his reconciliation with Elizabeth are mine, and so are most of the indications that Charlotte and her husband are heterosexual rather than asexual. The “young olive branch” is also mine, although chronologically this is several months before the subject comes up (in a letter to Mr. Bennet) in P&P. The smuggler’s run was something Claude and I overlooked in the initial outlining. It was the last thing drafted, but still required as much editing as the other chapters.

I was not entirely happy with even that version, so I turned to the “Writer Room” concept, where you prompt an AI to focus on different issues in different chats. In this case, I had the Claude chatbot generate the system prompts for each focus/persona I needed and then I collected them into one doc which I put in the “Files” section of a new Claude project. I then started three chats, told the chatbot which set of instructions to use inside each chat, and renamed the chats to reflect the persona and the order in which I expected to use them.

All three were designed to quote the problem passage, name the issue, and suggest a direction for correction without hedging. Since this was just free content for my website, I had no qualms about asking them to draft replacements for the problem passages and integrate them into the text I’d fed them. They generally came up with solid ideas when that became necessary. In the past, I’d mildly scoffed at the people who give genre appropriate names to these personas/functions, but I found myself doing the same, or least asking Claude for ideas of what to call them and picking out the ones I liked. Here they are in the order I used when I cycled through them in each chapter.

1). The Intelligencer covered character behavior and dialogue authenticity, across two categories: canonical Austen characters who needed to stay faithful to the source, and the reimagined or original characters who needed to stay consistent with the versions this story had established. In theory, it flagged anything that felt plot-convenient rather than character-driven. In practice, it offered some minor tweaks to scenes where I had softened the personalities of Charlotte or her husband more than appropriate, and caught a few typos.

2). The Riding Officer covered setting, atmosphere, and sensory grounding. In my own writing, I fight constantly against the tendency towards whitespace (characters talking in a void with no setting) and for whatever reason Claude had been guilty of the same in its original draft of this fanfic. The story needed to balance two worlds simultaneously: the sunlit Regency surface of parsonages and dinner parties, and the nocturnal smuggling world beneath it. The Riding Officer was designed to address this, and it did more work than the other two, in terms of finding issues and being asked to draft solutions. It made its share of mistakes, like the time it conflated the front grounds of Hunsford with the front grounds of Longbourn (last seen several chapters before that point). Already with only 15000-20000 words of manuscript to discuss, it was starting to lose coherence. There were also a couple times when it came up with paragraphs that repeated material that was already there. But it corrected its mistakes when I pointed them out, so the time cost for me was pretty minor. It also caught and (mostly) fixed some time-of-day continuity issues in Chapter 16.

3). The Raconteur was supposed to keep an eye on narrator voice and prose consistency. Even when given an example of my own first-person prose and told to use this as a benchmark for voice, it stuck stubbornly to the style originally established by its fellow Claude in the chat where the fanfic was originally drafted. Its actual work consisted of correcting a few typos the Intelligencer overlooked, and cleaning up some overwritten or redundant passages the Riding Officer had introduced into the text. There were also cases where it simply reverted to an earlier version of the text rather than integrate changes properly, so Chapters 8 and 9 are just the Riding Officer’s version with some minor tweaks by me.

If I were doing something like this part for a work of original fiction, I would probably look at the AI’s prose suggestions as prompts for me to work on my own writing rather than additions I would casually accept. I would probably drop the Raconteur/prose consistency persona in favor of a copyeditor type persona, and work longer and harder at developing the character consistency persona, because in this case a lot of the work had already been done for me by Jane Austen. As with everything else about the chatbot AIs, you need to figure out what you want from them before you put them to work.