The Conversation Loop: Why Your Best AI Writing Comes from Iteration

A practical tip for steering AI output until it sounds like you

I see it constantly in author communities: someone opens up ChatGPT or Claude, types in a prompt, gets something back that feels generic, and immediately declares the whole thing a waste of time. “AI writing is slop,” they say. “It doesn’t sound like me. It doesn’t sound like anyone. What’s even the point?”

And honestly? I get the frustration. That first output usually is pretty generic. But most people miss that the first output was never supposed to be the finished product. It was supposed to be the start of a conversation.

The single biggest AI writing tip I can give you is this: treat AI like a conversation, not a vending machine. You don’t put in a quarter, get a chapter, and walk away. You talk to it. You react. You push back. You refine. And by the end of that process, the output doesn’t sound like AI anymore. It sounds like you.

I call this the Conversation Loop, and it’s the difference between authors who love working with AI and authors who tried it once and gave up.

The Moment Most People Quit

The problem is not that the AI “can’t write.” The problem is that the author didn’t tell it what “good” looks like. They expected the AI to read their mind — to intuit their tone, their pacing preferences, their character’s inner voice, the specific emotional register they were going for. And when the AI didn’t magically deliver all of that from a single sentence of instruction, they concluded it was broken.

This is the same reason I wrote my earlier piece, The AI Can’t Read Your Mind. Authors are still, somehow, surprised that a tool designed to respond to input can only be as good as the input it receives. I don’t know why this expectation persists, but it does, and it’s the number one reason people abandon AI before they’ve given it a real shot.

“Revise the Scene” Isn’t a Prompt

Let me give you a concrete example of what I mean. When you tell an AI to “revise the scene,” you haven’t actually said anything useful. Revise it how? More emotional? Faster pacing? More internal dialogue? More sensory detail? A clearer inciting incident? Shorter sentences? Different POV? You could mean any of those things, and the AI has no way of knowing which one you’re after.

A human editor would ask follow-up questions. They’d say, “What’s not working for you? Is it the pacing or the tone? Are you looking for more interiority here?” AI doesn’t do that — at least not unless you’ve specifically trained it to. So you have to do what you’d do with a human collaborator: you have to talk back.

That’s the fundamental mindset shift. AI isn’t a search engine where you type a query and get an answer. It’s a conversation partner. And like any conversation, the quality of what you get out depends entirely on the quality of what you put in — and your willingness to keep the conversation going.

The Conversation Loop

Here’s how the loop works in practice. You generate a draft. You read it and react — not with “I don’t like it,” but with specifics about what’s missing or what’s off. You specify what to change using real craft language. You regenerate. You refine. And you repeat until you hit that moment where you think, “Yep, that’s it.”

This is not extra work on top of the writing process. This is the writing process. It’s the same thing you’d do if you were workshopping a scene with a critique partner or going back and forth with an editor. The only difference is that the feedback loop is faster and you can iterate as many times as you want without anyone getting annoyed.

The number of rounds it takes depends on how much setup you’ve done. If you’re starting cold — just a raw prompt with no context — it might take five to eight rounds to get something you’re genuinely excited about. But if you’ve done the legwork of providing a style guide, sharing example text for the AI to reference, and having an upfront conversation about the kind of output you’re looking for, you can get there in three or four rounds. The setup pays for itself.

What This Looks Like: My Cozy Mystery Test

One of my go-to test prompts when I’m evaluating a new AI tool looks like this: “Please write Chapter 1 of a cozy mystery novel where a woman is out walking with her dog, the dog gets off the leash, runs away, and finds a dead body in the bushes.”

If you’ve used AI at all, you can probably predict what happens next. The AI will give you something passable. Maybe even a little funny. But it will also be utterly generic, because I didn’t do the legwork of telling it how to write this scene.

Think about everything that’s missing from that prompt. I didn’t mention the woman’s name, the dog’s name, or the town they live in. I didn’t specify what mood I wanted, what POV to use, whether it should be first person or third person, what time of day it is, whether there are other people around, how long the chapter should be, or what kind of voice I’m going for — cozy and quippy? Cozy and lyrical? Something in between? I didn’t tell the AI what kinds of details I like to linger on or what sentence structures I prefer. All of that context is just... absent.

That’s not an AI problem. That’s an input problem. And this is exactly where the Conversation Loop kicks in. I’ll start adding constraints. I’ll tell the AI about the character and the setting. I’ll request specific craft moves. I’ll reshape the tone. I’ll tighten the pacing. Each round gets closer to what I actually want, because each round gives the AI more information about what “good” looks like to me.

Nobody Is Just Pressing a Button

Every time someone argues that authors are “just accepting AI output and publishing it,” I have to laugh a little. That’s really not what’s happening.

Most authors I know who use AI seriously are not accepting first outputs. Even the people who do lean toward a “generate and accept” approach usually do it only after they’ve put in significant setup work. They have a style guide in place. They’ve provided example writing for the AI to mimic. They’ve had conversations — sometimes long ones — about the kind of output they want. And even after all of that, the output still gets edited because it’s going into a larger work that has its own rhythm and continuity.

“Generate and accept” is really more like “generate and lightly edit” — and it happens after a ton of front-loaded work that nobody sees. The idea that people are typing “write me a book” and uploading whatever comes back is a fantasy that makes for good outrage on social media but doesn’t reflect what’s actually happening in the community.

Speak in Craft, Not Vibes

The fastest way to get better AI output is to stop speaking in vague vibes and start speaking in craft. This is probably the most practical thing I can tell you.

When I’m steering AI mid-conversation, I use specific writing craft terms. I’ll say things like “add more visceral interiority and internal dialogue here” or “this needs to function as an inciting incident — raise the stakes” or “use shorter, punchier sentences to increase the pace of this scene.” I might ask it to shift to a closer POV, or to make the subtext clearer without making the dialogue on-the-nose.

Notice what all of these have in common: they tell the AI what to do and why. The more specific your craft vocabulary, the fewer rounds you need. If you can name the problem, the AI can usually fix it. But if all you can say is “I don’t like it” or “make it better,” you’re going to go in circles. Investing in your understanding of writing craft — things like POV, pacing, interiority, dialogue mechanics, scene structure — pays enormous dividends when you’re working with AI. The craft knowledge becomes your steering wheel.

Knowing When It’s Good Enough

This part is entirely up to the author, and honestly, it’s a gut thing. I know an output is good enough when it meets my expectations, when it sounds like me, and when my instinct says, “Yes. I can work with this.” That’s the threshold. Not perfection — just the point where the text is ready to be folded into the larger manuscript and refined in the editing process.

That’s easier for me because I already know my voice. I wrote over 40 books before I ever used AI, so slotting AI-generated text into my own work happens naturally because I know exactly what I’m looking for. But this is true for other authors too, even newer ones. The more you write and the more you read, the sharper your sense of what “sounds right” becomes. AI doesn’t replace that instinct. It relies on it.

The Output Becomes More You with Every Round

This is the part that surprises people, and it’s really the heart of what I want you to take away from this article.

AI is a conversation. Just like talking with another author, an editor, or a critique partner, you’re narrowing the result using your own judgment, curation, and taste. Your preferences are the compass. Your voice is the destination. And after a few rounds of the Conversation Loop, the output stops feeling like “AI writing” and starts feeling like your writing — because it is your writing. You shaped it. You steered it. You decided what stayed and what got cut. The final product is more you than anything else.

That’s the reframe I want to offer: AI doesn’t write for you. It writes with you. And the more you engage with it — the more you push back, specify, and refine — the more the output reflects who you are as a writer.

Try This Before You Quit

If you’ve tried AI once and hated it, I’m going to ask you to do one thing before you declare it hopeless: run three rounds of the Conversation Loop. Generate something. React to it with specifics. Tell the AI what to change and why. Regenerate. Refine. Do that at least three times, and then decide whether AI writing is “slop.”

Because in my experience, most people don’t hate AI writing. They hate the first draft. And honestly? Welcome to writing. First drafts have always been rough. (I’ve never been a fan.) The magic has always been in the revision. AI just makes the revision loop faster.

So have the conversation. Push back. Be specific. And see what happens when you stop treating AI like a vending machine and start treating it like a collaborator.

You might be surprised by what comes out the other side. 😉

---

Have you found the Conversation Loop helpful, or are you still getting stuck at that first output? I’d love to hear what you’ve tried in the comments.