My most recent is “The Silent Fjord”, a Viking-inspired murder mystery. I say inspired because I was too lazy to do much historical research. Older YA/New Adult. It started as an exercise in figuring out how to use Antigravity but I gave up because Gemini is terrible at plotting. I then took the concept to Claude snd we refined it, then into Novelcrafter where ChatGPT 5.1 got the book done. ChatGPT generated the cover but it needs a new one because it looks ok large but gets a bit dense as a thumbnail. Truth is dangerous. Magic is just another kind of lie. The Silent Fjord by Jade Kirkland.

It depends on what you mean by writing. AI generates 100% of my prose but the part where I sit in the sun with a glass of wine and dream up a new idea is still writing, and I do 100% of that. But generally speaking, it’s usually 100% my idea and 50/50 brainstorming. The AI does 75% of the world-building and codex development, but the outline is usually 75% mine because I’ve yet to find an AI that doesn't just write generic stories. Then it’s 100% AI prose, but it’s using a prompt that took me six months of tweaking to make write like me, so maybe I’m giving it too much credit for the words. Then editing is all me, but how much I have to do depends on which model I use. Which is a long-winded way of saying I voted 50% because it’s a collaboration.

Depends on the project, I think. I'm just stating. anew book and did my usual practice of getting my LLM contenders to write the first chapter or scene and see which one nails it. Opus 4.6 was ok but I had to edit a lot of foreshadowing, and explaining everything as if the reader is an idiot - the sort of stuff I haven't seen from an LLM in months. If I were to use Opus 4.6, I'd definitely have to tweak my prompts to compensate. But it wrote some good lines. And once edited, I'd have been happy to use it. Sonnet 4.5 did great, but my writing prompt in Novelcrafter has been slowly modified to work with it. I think the it nailed the characters and dialogue better but Opus was maybe slightly better at descriptions and humour. Between the two, my choice would be Sonnet 4.5, simply because, for the extra cost, Opus wasn't that much better. Out of 1,200 words, I had to edit maybe 15 lines with Opus vs only 6 with Sonnet. And just for the record, I ultimately went with chatGPT 5.1 yet again. It wrote somewhere between the two Claudes. Great snarky humour. Not quite as evocative in its language, but very distinct character voices, And there was only one edit to make in the whole scene.

@Gabrielle Brown I'm specifically using GPT 5.1 via the API. I haven't played much with 5.2 but it's 50% more expensive so I won't be using that until they take 5.1 away from me. But from what I've seen, If had to choose between GPT 5.2 and Claude, I'd probably go with Sonnet 4.5.

As Dana said, you should have some scene summaries in the write section that your chapters were based on, and the lower left button should still show the model and prompt you used to do the writing.

@Sarah Barbour The bottom left button in this screenshot will be set to whatever model and prompt you last used to create the prose based on those beats.

The weekend’s over but what I did was outline a short story for my Juniper Harlow paranormal mystery series to use as a giveaway for joining the mailing list and I got it 90% there. Juniper’s stories are usually good, snarky fun, but my goal for this one is to make the reader cry as well as laugh, and I haven’t quite nailed it yet. And I started designing some Facebook ads for the Viking era YA/New Adult murder mystery I released at the end of last week.