Transcription
Here are four prompting techniques that will make it so that your books will go from being obvious AI slop to being something that nobody would guess is actually written by AI. It's that good. I use these techniques heavily in my automations. I'm doubling down on all of them as I'm going through and creating my own tool for writing with AI. And this is, I believe, these four techniques, if you can figure this out, this is going to be extremely helpful for you in the writing process.
Now, most people are not using these techniques because, frankly, if you're using them inside of a chatbot or something like that, it might get a little tedious, and most people aren't into that. And that's why I use automations for a lot of these because it can run through several of these steps and techniques without me really having to supervise it. It just does it automatically. But you can absolutely use all of these in the process. So let's dive into these four techniques and figure out how can you level up your writing using all of these.
The first is by utilizing checks or critiques of your work. Now, this might sound simple, but a lot of people get this wrong. Let's say you've written a chapter using AI based on what you told it to do. There are going to be a lot of issues with that chapter right out of the gate. It's going to have a lot of AISMs in it. It's going to probably have some poor writing examples in there, like maybe too many adverbs or something like that. And you ideally would love to filter out as many of those things as possible so that you don't have to change it manually. And then when you go in to fix it up yourself, you're only focusing on those aspects of the world, the chronology, and just making it match your own vision as much as possible. And maybe just adding a little bit extra human flare to it that the AI wouldn't necessarily get, but we can at least minimize a whole lot of the issues the AI has. So you're not just playing whack-a-mole with all of these issues that you're actually starting with a fairly clean draft for you to work on.
The way this would work is you take your chapter and you would run a prompt on it that says something to the effect of, "I want you to search for all instances of superfluous adverbs and come up with an improvement plan to remove these adverbs." Now, this is the key point of a check like this. We're not going to ask it to rewrite anything. We're just going to ask it to create a plan on what it would change. This allows it to focus on a very narrow set of parameters. And this is key to the checks approach is that ideally, you want to focus on just a few things with each check. And thankfully, if you narrow it down enough, you can use a really inexpensive model like say, Gemini 3 Flash or one of those to do this check. Because you've narrowed it down, like let's say we're only focusing on adverbs, you could have a check that includes a giant list of a whole bunch of things that you wanted to watch out for, but that runs the risk of it getting too spread out in what it does. If we focus what it looks at, then we can do this. And then we can run multiple checks to fix this thing up.
So you ask it for an improvement plan, and then in a second step to this, you ask it to incorporate the feedback from the improvement plan but to change nothing else other than what was on the improvement plan. This is another key aspect to it because if you don't do that, it's likely to rewrite the entire thing and do it probably worse than the original draft was, especially if you're using a less powerful model in that redraft process. But if you just asked it to change what was in the improvement plan and leave everything else the same, you can use a lesser model to essentially improve your draft.
So, this is a technique that not enough people are using, partly because it takes quite a lot to go through all of these checks if you're just using a simple chatbot. Though, you can do it. This is why I do it inside my automations where I will just have a series of these steps that it goes through, and it just does them automatically. And you can have all kinds of checks. So you can have checks for adverbs, check for dialogue tags, checks for the logic and plausibility of the scene, checks for various types of AISMs. Like, you know that thing that AI does a lot where it says, "it's not just this, but it is also that." That's something the AI does quite a bit. You could create a check that specifically looks for that particular type of phrasing and gives you an alternative to say instead. And then you just run that check, and it filters out all of those things. It's a great way to utilize humanization in this whole thing.
So, that's the checks or critiques process and creating the improvement plan. It's one of the most powerful ways to improve your writing as long as you have a solid series of checks and prompts for each one to go through and filter these things out.
All right, the second key prompting technique that I think is much more important than people give it credit for is the value of context engineering. So, context engineering is this term that's come up recently. This idea of, it's not really about the prompt so much as the quality of the context that you give the AI. So, from a fiction writer's perspective, this is pretty obvious. This means, like, you give it all of the information about your world, about your plot, about your characters that you want it to know. If you don't give that information, it won't know. You know, it'll come up with something more generic that doesn't match your vision. But it actually goes a little bit beyond this. For instance, if you're writing a chapter and you're providing it with a whole bunch of information, it might pick the wrong bits of information to write that chapter. Instead, we kind of want to curate what gets fed into the AI's input when it's writing this chapter. So, maybe only the characters that are in that scene, or just the key details that it needs to know for that scene. And if you're using a tool like NovelCrafter, NovelCrafter actually is excellent at managing your context and making sure that only the little bits of information that the AI needs to know for that scene that you're writing is the stuff that the AI knows. NovelCrafter can be a little bit of a complex platform, but it is excellent at context engineering.
If you're running, say, an automation, which is how I do like 90% of my AI writing these days, I would put some steps in there that kind of look at what needs to be written and then looks at the total character sheet and all of that and just selects the pieces that need to be used in that particular scene. And then those pieces are the ones that are passed on when writing the chapter.
Another key aspect of context engineering that you need to keep in mind is not giving it too much context. This is not just because you don't want to give it information it doesn't need, but also because this can actually water down the prompt. I've seen too many examples of people putting their entire first book into the context while they're writing their second book, or even if they're writing just the first book, they might have the entirety of everything that's been written up to that point available in the context as they're writing the next chapter. And I strongly advise you don't do this because what you're doing is you're watering down the prompt. You're making it so that if it needs to access certain details, it's like a needle in a haystack.
So, part of context engineering is also providing the AI with succinct and important summaries of the text that it needs to know. This is why I actually have an automation that I use called creating a summary plus document. One that scans through your entire book and summarizes each chapter one by one with just the key details from that scene, like it gives a one-paragraph summary of the scene. It gives the list of characters that are in the scene and any descriptors of that character, what happens to them, and a few other details like that. And it condenses the amount of text that is used, or it condenses the entire book into something that's much more manageable for the AI to understand. And it also signposts everything so the AI can quickly find what it's looking for, which is much better than just trying to dump the whole book in there and just hoping that it knows everything that's in the book, which even though it's supposedly in its context window, that's not really how it works. The bigger the context, the more it has to sift through, the worse it gets at actually identifying relevant details that it needs for the scene. So, this is why I tell people that you should try to minimize your context. Try to keep it under 50,000 tokens if you can because that's just going to result in a higher quality output because of that.
Quick pause. If you're an author who wants to start writing more books, I put together my complete AI prompt pack that helped me and many others in my community write a book in a month. It's completely free. Hit the link in the description. Join the free waitlist for StoryHacker Gold. Then you just answer a couple quick questions, and I'll send you the full prompt list plus a few extra free goodies. I'll also email you when I open the paid StoryHacker Gold group so you can join if it feels right. There's no commitment to join, though, and you'll still be able to get all of the freebies you signed up for either way.
All right, back to the video.
All right, the third major prompting technique is one I'm kind of excited about and kind of a new one, too. This was first Future Fiction Academy came up with this idea that they call narrative physics. And I've expanded on it a little bit to where I think of it as sliders, or you can think of it as game stats for different aspects of your world. Now, this is still just a prompting technique. There's really nothing crazy about it, but I'll give you an example of how this looks. And in the future, I'll probably make a whole video getting deep on one or more of these techniques, but here's the idea.
If you have a character, let's say you have a character bio for a character that's in your book, and you include that character bio in every scene that that character shows up in. Well, what if that character changes across the course of the story? And what if their emotions are in the same in a different place so that they react differently? Like, you might have a character who is normally a very nice person, but under a massive amount of stress might break and like kill somebody, you know, like under the right circumstances. I feel like there's a joke there from like The Dark Knight and the Joker that if you just give people a little push, they'll do anything, right? That's this idea that people behave differently in different situations. And so what I came up with, and this while the concept of these narrative physics is something that Future Fiction Academy came up with, I have not seen any of their prompts. I have no idea how they do what they do. I came up with my own based on their concept. And the way I came up with it was this idea of sliders.
Um, so for every character, I will now give them sliders for things like stress to harmony or peace, right? And if they're like a number 10 on the harmony side, there might actually be like, too harmonious and a little bit too out of it, like maybe they're on drugs or something. But if they go all the way to the other side of the slider, like a -10 on the stress side, they're like so stressed, they're having a panic attack, they're literally on the verge of dying. Whereas we can pick somewhere in the middle is where their baseline is. And then for each scene, we can get the AI to identify, okay, where are they still at their baseline, or have they moved up or down from that baseline? And what this does is it allows the characters to shift and react more realistically to the scene. And so this is an excellent technique to get a little bit more variation about your character so they don't always sound monotone and the same all the time.
It's also an excellent technique if you use something similar for, say, you have a magic system, right? How does what are the rules of that magic system? How does it work? What are the costs? All of that. Making sure you establish all of those things in almost a computer-like or game-like way with actual numbered stats and things like that that can then change and evolve over the course of the book. You can even do this with certain emotions that you want to get across inside of the story, like the amount of tension that's in a scene. You know, what's that look like on a slider, right? Or the amount of spiciness in a scene. What's that look like on a slider, right? Sometimes you're going to have zero spiciness because the characters hate each other or something like that. Sometimes you're going to have, like in a romance, you'll have it go up and down, up and down, and up and down, but slowly getting bigger and bigger, and then suddenly, like, bam, right? You can take it all the way to a number 10, which would be like pornographic erotica, right? So that's the kind of thing we're talking about here is different sliders, different ways of telling the AI, this is the sort of tension that we're looking for. This is where the character is at in terms of their own sliders and so on. And so this is a really cool technique that I'm excited to be playing with more in the future.
All right, the last technique that I really want to talk about is, for lack of a better term, I'm just calling it layering. All right. And this is the idea that we can actually utilize multiple strengths of multiple different LLMs in one batch. And rather than just choosing one LLM to give us everything that we want it to give us, we can take it in steps and use different LLMs for each step. This not only allows us to play to the strengths of the LLM, but it also allows us to kind of avoid some of the sameness that we get from any one LLM. For instance, I love the Claude models. I love Claude 4.6 Opus, which is the most recent one as I'm recording this. And but when I work with it too long, it starts to all sound a very similar kind of tone all the way through. This layering technique allows us to mix it up a little bit and make it a little bit better.
So, what would this look like in practice? So, for instance, the Gemini and the ChatGPT models tend to be really good at logical thinking and structuring things. And so maybe we could start out by having them look at the information we provided about the scene and say, "Okay, let's expand this and create a scene brief." And so it creates the scene brief along with all the little things that we need it to have. And then, we know Grok is really good at dialogue. So then maybe we could pass the scene brief over to Grok and be like, "Can you create a dialogue-only draft of this and just put in little placeholders with a little bit of blocking for what happens in between the dialogue?" And then from there, we pull it over to, let's say, Gemini. And Gemini is pretty good at prose overall, but not as good at other things. So we can say, "Write all the prose, you know, fill in all the prose for this. Identify the chill scenes and write all the prose for this stuff." And then we pass it over to Claude Opus 4.6 and be like, "You're really good at writing really serious, dramatic, climactic scenes. Go through anything that's left in this scene and write in those little bits and bring it all together." So that now you have a chapter that's been gone over by like four or five different LLMs, all providing what makes them unique to them, and you end up with a chapter that is absolutely mind-blowing compared to what you would get if you just went to Claude or Gemini and asked it to write the whole chapter in one go.
This is a really powerful technique that I think a lot of people aren't utilizing. It does admittedly increase your costs of being able to do this, especially if you're using an API. It's going to take more, especially if you're doing this in like an automation type of format. It's going to take a lot more money to do that. But if that isn't an issue for you, this could be an incredible way to get really, really good results out of your LLM, especially if you combine it with the context engineering and the checks and critiques and the narrative physics and the sliders. You combine all of these together, you're going to get some pretty fantastic results.
So, these are some four pretty advanced tools that I think are really helpful for you if you want to get access to a lot of my stuff, including all of my automations that get into this sort of thing that I'm currently building as we speak. You can access the StoryHacker Gold group down below. And if it's on the waiting list when you join, you'll actually get a whole bunch of my prompts all for free that you'll that'll just be sent to your email when you join the waitlist. And then you can, you know, join the StoryHacker Gold group when it opens more publicly.