AI is a tool to master, not an expert to obey.

There is an entire industry devoted to teaching you how to talk to AI.

Use this prompt. Copy this template. Give it a role. Add these seven instructions. Tell it to act as a bestselling novelist, developmental editor and professor of creative writing with 30 years of experience.

Then paste in your chapter and wait for wisdom.

This is probably one of the least interesting ways a writer can use AI.

Not because prompting doesn’t matter. Clear instructions matter with any tool. But the obsession with finding the perfect prompt misunderstands what makes AI unusually useful to a writer.

You can talk back to it.

You can bring it a problem you haven’t solved yet. You can give it your theory. It can give you another one. You can reject that theory. It can ask you a question. You can realize you’ve been thinking about the wrong character entirely.

The useful part isn’t necessarily the answer. It’s the conversation that gets you somewhere you weren’t before.

AI is software, not your editor

Think about Excel.

Excel can perform calculations in seconds that would take you considerably longer by hand. It can organize enormous amounts of information, model different scenarios and reveal patterns that would otherwise be difficult to see.

That doesn’t make Excel a financial expert.

Give a powerful spreadsheet to someone who doesn’t understand the numbers and you have a powerful spreadsheet operated by someone who doesn’t understand the numbers.

AI isn’t fundamentally exempt from this rule.

The difference is that Excel doesn’t talk like it knows what it’s doing.

AI does.

Ask it what’s wrong with Chapter 12 and it will tell you. Ask it how to fix Peter and it will give you five suggestions. Ask whether you should kill Wendy in the third act and it can produce a beautifully organized argument for killing Wendy.

Then ask it why keeping Wendy alive is the better choice, and it can give you a beautifully organized argument for that too.

That doesn’t make the tool useless. It tells you what your job is.

You have the expertise. AI has capabilities.

Learning to use AI well means figuring out how to combine the two without confusing one for the other.

Don’t ask AI to have your opinion for you

Imagine you’ve just read a chapter of a novel and something isn’t working.

There are two ways you could approach AI.

You could paste in the chapter and ask:

Analyze this chapter. Identify its problems and suggest improvements.

You’ll probably get an answer. It may even contain useful observations.

But you’ve immediately handed over the most important part of the process: deciding what you think.

There’s another way: start with your reaction.

I didn’t care about this chapter until the argument between Peter and Wendy. Everything before that felt flat. I think part of the problem is that I’m meeting too many characters before I have a reason to care about any of them. But I’m not sure that’s the whole problem. Help me figure out why the argument works when the earlier scenes don’t.

Now you have something to work with.

AI can compare the scenes. It can challenge your explanation. Maybe it notices that the argument is the first time a character desperately wants something another character refuses to give him.

So you push further.

Is that really the difference? There was conflict in the earlier scene too.

Maybe the earlier conflict didn’t carry a meaningful consequence.

Okay, but I don’t think stakes are the whole problem either.

Good. Keep going.

By the end of the conversation, you might decide your original diagnosis was right. You might discover the problem isn’t the number of characters but the fact that none of them has been given an immediate desire. You might decide the chapter isn’t broken at all and the argument simply reveals what the rest of the book needs more of.

Or AI might give you a completely stupid explanation.

That’s useful too, provided you know enough to recognize it.

Because explaining why it’s wrong forces you to define what you think is right.

Bring AI your half-formed thoughts

Writers spend a lot of time having thoughts that aren’t ready to become thoughts yet.

Something is off here. Peter feels different in this chapter. I like this scene more than the last three and I don’t know why. The middle is dragging, but cutting anything feels wrong. Wendy’s decision makes sense logically, but I don’t believe she’d make it.

Those are excellent places to start.

You don’t need to convert them into a sophisticated prompt first. In fact, doing that can be counterproductive. If you force yourself to define the problem before the conversation begins, you may bake the wrong diagnosis into the question.

Just bring the mess.

Explain what you’re noticing. Give the AI the relevant material. Ask what it sees. Then interrogate what comes back.

That’s much closer to talking through a manuscript with another person than operating a vending machine where sufficiently precise instructions dispense the correct answer.

The goal isn’t:

PROMPT → ANSWER

It’s:

PROBLEM → QUESTIONS → DIAGNOSIS → OPTIONS → WRITER DECIDES

That last part matters most.

Make it show its work

Suppose AI tells you Peter’s motivation isn’t clear.

Don’t immediately fix Peter’s motivation. Ask why.

What specifically in the scene led to that conclusion? Where does his behavior become inconsistent? Is the problem that his motivation isn’t clear, or that his motivation is clear but not convincing? What evidence supports the distinction?

Now you’re examining the analysis instead of accepting it.

You can also make AI argue against itself.

Give me the strongest case that this scene should be cut.

Then:

Now make the strongest case for keeping it.

Or give it your own diagnosis:

I think this scene drags because Wendy gets what she wants too easily. Test that theory. What evidence supports it, and what evidence contradicts it?

That’s a much more interesting use of the technology than asking it to rewrite the scene.

You’re using AI to generate intellectual friction.

And friction is useful because writers are exceptionally good at becoming accustomed to their own choices. After you’ve lived with a book for a year, something can feel inevitable simply because you’ve read it 40 times.

A tool that can keep questioning the inevitable is valuable.

You just shouldn’t mistake its questions for verdicts.

You don’t need AI to agree with you

This may be the most important habit to learn.

Push back. Say:

No. Peter isn’t afraid of Wendy leaving. He’s afraid she’ll stay for the wrong reason.

Or:

You’re overemphasizing the plot. My problem is that I don’t believe the emotion.

Or simply:

I don’t buy that. Give me another explanation.

You aren’t breaking the AI when you disagree with it. You’re using it.

Sometimes the most useful moment in an AI conversation is the moment you become irritated because it keeps misunderstanding what you’re trying to accomplish.

Because then you type:

No. That’s not what this scene is about. This scene is about Peter realizing that Wendy has already made the decision and he’s the last person in the room to understand it.

Well, there it is. You knew what the scene was about. You just hadn’t said it yet.

The better you get at writing, the more useful the tool becomes

This is where the conversation about AI and writing often gets backward.

AI can make sophisticated tools available to inexperienced writers, but it doesn’t eliminate the value of learning the craft. It increases it.

If you understand point of view, you can interrogate a point-of-view problem.

If you understand scene construction, you can test why a scene isn’t turning.

If you’ve developed an ear for dialogue, you can recognize when an AI suggestion technically solves a problem while making the dialogue worse.

If you’ve read widely enough to have taste, you can reject the competent, obvious solution in favor of the stranger one that belongs in your book.

The person operating the tool still matters.

A great photographer can do more with Photoshop than someone who has never learned composition. A financial analyst can do more with Excel than someone who doesn’t understand a balance sheet.

The same principle applies here.

Learning AI doesn’t replace learning to write.

Learning to write gives you more to do with AI.

Stop collecting prompts. Start learning the tool.

Keep useful prompts if you have them. Build workflows. Give AI context. Tell it what kind of feedback you want. Learn which instructions reliably produce useful results.

Those are all part of mastering the software.

But don’t confuse a collection of prompts with mastery.

Mastery is knowing when to ask for an answer and when to ask for questions.

It’s knowing when more context will help and when the AI is getting distracted by too much of it.

It’s knowing when to ask for five alternatives, when to make it defend one, and when to ignore all five.

It’s recognizing when the tool has spotted something you missed.

And it’s recognizing when the tool is full of shit.

You don’t need AI to write your novel for you.

You need to become good enough at using it that it helps you see your novel more clearly.

Bring it the problem. Talk through it. Argue with it. Change your mind when it earns it. Ignore it when it doesn’t.

AI isn’t the expert. You are.