Don't Start Yet, I'm Still Feeding You Info: How I Actually Talk to AI

Here's a scene I've watched play out a hundred times. Someone opens a chat, pastes in three paragraphs of half-formed thoughts about a new offer they're launching, hits enter, and waits. The model, doing exactly what it's built to do, takes that as a green light. It drafts. Confidently. Immediately. And what comes back is polished, well-organized, and completely wrong, because it guessed at the parts nobody told it.

I'm Zora, and I want to show you the fix, because it's one sentence, and almost nobody uses it.

The Draft Nobody Asked For

Picture a small business owner, let's call her building a new service page. She opens a chat and types out everything she can think of: her pricing, a little about her background, a sentence about who she serves, a half-formed idea about a guarantee she's not sure she wants to offer yet. Then she hits enter.

The model doesn't know she's not done. It sees a wall of text and treats it as a brief. Thirty seconds later she's staring at a full draft, guarantee included, tone confident, structure clean, and entirely built on the parts of her brain dump that were never meant to be final. Now she's not refining a draft. She's untangling one. That's a slower path than starting clean would've been.

The One Sentence That Changes It

Compare that to what happens when you say this first: "Don't start yet, I'm still feeding you info." Four words that do more work than any clever prompt you'll find on a list somewhere.

Here's what it buys you. The model holds. You keep dumping your messy, half-formed, out-of-order thinking into the chat, because that's how your brain actually works and you shouldn't have to perform tidiness for a machine. When you're done, you say so. "Okay, that's everything." Now the draft it gives you is built on the whole picture instead of the first third of it.

I use a few versions of this depending on what I need. "Ask me follow-up questions before you draft" when I want the model to catch gaps I didn't notice. "Only use what I provide, don't guess or hallucinate" when I need it to stay inside the facts I gave it instead of filling silence with something plausible-sounding. And two newer ones that have earned a permanent spot: "hold for more information" mid-conversation when a new detail just occurred to me, and "audit your work" or "check your work before you come back to me" when I want it to catch its own mistakes before I have to.

Watching the Difference

Run the same project two ways and you'll see it. Skip the expectation-setting step and you get a draft that's confident, clean, and wrong in ways that take real time to unwind. Set expectations first and you get something slower to arrive but far closer to right, because the model wasn't guessing at your intent, it was working from it.

This is step three of the AI DNA Workflow for a reason. It sits right after you've defined your outcome and chosen how you want to feed the AI, and right before you actually hand over your context. Skip it and every step after gets built on a shaky guess. Do it and the model becomes something closer to a collaborator who knows to wait for you.

Say It Before You Paste Anything

Next time you open a chat with something half-formed in your head, say the sentence first. "Don't start yet, I'm still feeding you info." Let yourself be messy. Let the model wait. You'll spend less time untangling a draft you never asked for, and more time actually building the one you meant to make.