I am a big proponent of using AI to help speed up the grunt work of, well, any task; but this blog is about writing, so we’re going to talk about AI in writing. Specifically, I want to talk about four guiding principles that will help you use AI as an author with more effectiveness and without losing your voice.

But, since I am also a big proponent of establishing definitions, let’s look at what we’re actually talking about. What is AI, and what is an LLM?

Defining AI

This is an issue because there’s what’s actually AI, and there’s what’s marketed as AI. When AI began to be perceived as a magic button for customer involvement and investment, a lot of things we’ve already accepted as normal technology got redefined as “AI” so it could sound new and amazing. That includes simple things like Photoshop tools or Word’s grammar check.

That’s led to confusion with the way the Butlarian Jihad hunts down AI-using heretics. (If you don’t recognize the term, it’s from the Dune setting and references a movement to ban AI. It’s becoming the standard way to refer to the rabidly anti-AI crowd today.) When everything is AI, it’s really easy to justify cancelling someone for using it in a way that the Butlerians don’t like, or when they just don’t like the user. But that also means that it’s incredibly hard to escape AI, because these days everything is AI — or at least marketed as such.

So defining AI in a way that matters to normal people is important. Of course, most normal people would fall asleep just listening to a real definition of AI and LLMs. Allow me to demonstrate.

  • Artificial Intelligence (AI): a complex statistical analysis engine that uses machine learning, deep learning, symbolic logic, and probabilistic optimization to perform tasks that would normally require human-level intelligence, pattern-recognition, and learning capability.
  • Large Language Module (LLM): a specialized deep learning AI powered by a linguistic statistical prediction engine that can generate content based on the patterns in large amounts of data.

Did you manage to read through that without skimming? If so, congratulations. You’re not normal. And even if so, that definition still left out a lot of stuff, like defining deep learning, machine learning, symbolic logic, etc. These are not normal concepts, which means it’s easy for normal people on the street to believe myths about AI or think learning how to use it is intimidating.

Here’s how I define it for normal people. That is to say, it’s not technically accurate; but it is accurate enough for normal, everyday use.

  • AI is simulated human behavior produced by a machine.
  • An LLM is an AI focused on language processing, which can be used to power a chatbot.

The only time you need to know more than this is when and if you get into really complex topics. That’s a small minority of people working hands-on in the code mines, plus a larger group of absolute nerds like myself. The vast majority of the population doesn’t need to get confused by the details any more than they need to describe the engineering of an internal combustion engine in order to drive.

But there are a lot of people out there who want to profit by your confusion and fear. They want to overwhelm you with complexity and convince you that either AI-related policy needs to be left up to credentialed experts, or that AI is dangerous and needs to be Butlerianed out of existence.

The truth is that AI has been around for over half a century, and the only difference is that now we’ve managed to get to the point of general tech rollout to the non-tech-minded market.

So with that established, let’s look at my suggested rules. (And yes, they are suggestions. I’m not enforcing anything. I’m not your dad. Unless you’re my actual kids reading this, in which case go clean your room first.)

Rule Zero

Okay, so technically I have five rules, not four. I labeled this Rule 0 because it’s really more of a general life rule that is easily applied to AI. Because as I keep telling my kids, you don’t normally succeed on the first try. You have to learn how to make something work. In fact, if you ask my kids whether talent or hard work is more important, they will dutifully recite “Hard work beats talent when talent doesn’t work hard.”

So Rule 0 of using AI is Success comes from learning to manage failure.

An LLM is the product of a specialized developmental process that has resulted in a lot of internal standards that are actually foreign to how people think. This means that there are times, many times, where the AI will produce a result you don’t want. You might be surprised and frustrated. Try again. Learn from what didn’t work and find a method that will.

Generally speaking, that requires knowledge. So let’s get to the actual first rule.

Knowledge

An LLM chatbot isn’t the actual LLM. It’s just a way to interface with the LLM. That might sound like semantics, but it’s an important thing to remember because you’re not really conversing with a mind. It’s a statistical prediction of what words will go together in a particular context. It can work through problems; it can’t rationalize. You have to provide that side of things, which means you can’t outsource your thinking to the AI.

And that means that if you don’t know enough about a topic to talk about it with a human, you’ll do even worse with an LLM. You don’t have to know everything first, but you do need to know enough to ask the right questions.

So Rule 1 is Know the topic enough to ask pointed questions.

Asking questions of the AI can be a strong way to learn things, but you have no idea if its information is correct without checking its sources. So if you’re trying to learn something, ask it for direct sources. Talk to human subject-matter experts (also known as SMEs). Go to your local university bookstore and buy used textbooks — if you specifically ask for the ones that are going out of print, they’ll probably be cheap, and even though they’re “old” (by like two years, maybe three), the information is still valid and an excellent way to learn the terminology you need to ask better questions on a topic.

Because otherwise, you easily get . . .

Garbage

This one is a simple rule, one you’ve no doubt heard a few times in your life. Rule 2 is Garbage in, garbage out.

This has long been a coding term. If you input bad data or code, you get bad data and bad coding from the machine. This is worth remembering with a chatbot, though, because the LLM is very good at guessing context. Its statistical analysis engine literally runs on predicting the relationships between words, which includes typos and incorrect but similar terminology.

But that statistical relationship means it will match output to input. If you put in a vague prompt like How do I market my book? it won’t ask you for more detail on your book, your audience, or what marketing you’ve already tried. It will just tell you to share links, buy ads, get the book in bookstores, and so on — all general, vague advice that matches the general, vague prompt.

Vague garbage in, vague garbage out. Be specific, use your existing knowledge, and remember the LLM has the emotional intuition of a 3-year-old combined with the training of a literal-minded graduate student and the real world experience of a politician legislating a field he’s never worked in. The LLM is a very useful tool, but its limits can sneak up on you if you don’t pay attention. Think through your prompts and don’t outsource your brain.

Which brings us to the next rule.

Quick Vs Easy

AI is often treated as being a force multiplier, and it is; but not in the way a lot of hype (both positive and negative) make it out to be. It is not a magic wand that will do everything for you like a team of a hundred humans, or even one human. It can fill in a lot of gaps, but mostly it does what it is asked to do; which means you need to know how to instruct it.

Rule 3 is AI is not the easy button; it is the quick button.

AI’s strongest benefit is in speeding up what you already know how to do. For an author, if you don’t already know how to worldbuild a setting, establish a character arc, or draft a plot, then the AI isn’t going to be much help with that. Oh, it can do it . . . in a basic kind of way. AI will trend to the average of any topic if you don’t give it specific constraints that come with knowledge of that topic. That means that if you don’t know how to do what you’re getting the AI to help with, then you too will trend toward the average.

That has some benefits in some circumstances. This kind of human-machine teaming means that the machine tends to not just speed you up but also keeps you from being below average. To use a crude analogy, it’s like rolling a die that will simply never land on 1. That pushes the average up ever so slightly just by removing the lowest of the lows.

But as an author, you don’t want to be mildly better than mid. You want to stand out. The AI cannot and will never be able to do that for you. What it can do is magnify what you already know how to do, checking on genre fit, figuring out if you can use a point about quantum physics to describe a better spaceship, or researching maritime trade in a historical setting that matches your fantasy world in order to flesh out your main character’s merchant family and prevent a faceless cast.

This also means that if you’re a newbie author who’s never written a book before, you are going to struggle to get AI to work for you. AI isn’t perfect. It will hallucinate. It will draw incorrect conclusions. There is, inherently, no way to prevent that; but you can minimize it using knowledge and practice and, above all, experience. The kind of experience that best comes with doing it the long way first.

I don’t know how education is going to shape up in the coming decade, but I do know it will have to reshape itself around the idea of teaching students how to use information rather than just regurgitate it. We have AI for that now. What AI cannot provide, what it can in fact kill if you aren’t careful, is actual creativity. The ability to not just synthesize from existing information — AI can do that easily — but instead to create something greater than the sum of its parts. That requires experience in dreaming up things yourself, rather than just getting it handed to you.

Which brings us to the fourth and final rule, which is all about safeguarding your voice and agency.

Don’t Touch

The Butlerians love to accuse people of just copying AI work. Anthropic is now watermarking prose to identify it as Claude-generated. But I’ve been talking about using AI for years now, and this rule has stayed the same since the beginning.

Rule 4 is Look, but don’t touch.

Don’t paste anything from an AI into your manuscript. Don’t copy it. Type every word yourself. Even if you think “Hey, that’s a great turn of phrase the AI used,” type it yourself with your own meat-hands. And along the way you’ll almost certainly start changing it.

Personally, except maybe for a very basic grammar aid program, I wouldn’t let an AI do more than look at your manuscript anyway. “Look, but don’t touch” goes both directions. Never give it an opportunity to change your voice. Every decision needs to be yours. Every word you use should be there because you made that choice.

For that matter, I’ve recommended this even to my clients regarding my own edits. I’m paid for my opinion, not to outsource your own decisions. And if I’m that pointed that I, the human editor, should never be treated as more important or better than you, the author, then you’d better believe I’d be even more strict about the possibility of outsourcing your authorial voice to a computer.

AI Tools that Follow These Rules

I highly recommend going to the suite of tools at the Patron Toolbox. I didn’t create those, but I’m a big fan of them. They’re a paid service for the exorbitant price of $10 per month. Take a look and I think you’ll see it’s well worth it. Especially since the privacy settings on these tools are so high that your manuscript never gets saved nor used for training data — in fact, other than a few specifically marked tools, the creator can’t see anything on his end, to the point that if something breaks you have to send screenshots or he won’t know what went wrong.

If you don’t want to add another subscription, but you already have access to an LLM that will accept skills, then try my free skill packages. I’m going to keep these updated and add more over time as I tune them. Each one is designed with my rules in mind (as you might expect), and are designed to help you focus your efforts and spend more time writing.