This series is not organized in any particular way but offers small tips that might be useful.
There are many ways to approach writing with AI, and I've learned that the biggest mistake is treating it like a puppet whose every move needs controlling.
I usually come to AI with a clear idea of what I want to say. I don't type "make me a topical op-ed" and see what happens—I have my own ideas and vision. But here's the challenge: chatbots want to be helpful. They'll overstep easily, trying to take over everything when you just wanted assistance. Sometimes I've literally told them: "This is my op-ed. When you become human, you can write your own. Follow my lead."
This comes from how these models are trained: humans reward responses that seem "helpful," which often teaches the AI to overdo it. The result? An overeager assistant that reorganizes your entire desk when you only asked for a pen.
I've found two approaches that actually work:
First, you can write everything yourself and ask the AI to fix grammar and do minor cleanup. This keeps you in complete control, though even then, you might find it trying to "improve" your style when you just wanted typos fixed.
Second—and this is my preference—give it the outline, major points, and examples of what you want to communicate, then let it write. These systems write remarkably well, but if you micromanage too much, it doesn't work. You're better off going as far as possible with high-level instructions before diving into specifics. You can accomplish a surprising amount by staying conceptual: "make it more conversational," "consider this counterargument," "adjust for a skeptical audience." The AI handles these shifts without you dictating individual sentences.
Think of it like working with a colleague. If your boss stands over your shoulder, constantly interrupting with corrections, the work suffers. You can't find your flow. The same happens when you micromanage AI—the prose becomes choppy and inconsistent.
The messy middle ground—trying to control each paragraph while the AI writes—is where things break down. You need to be either the driver or the passenger, not someone grabbing the wheel while someone else is driving.
Letting the AI take a first pass isn't giving up control—it's just sequencing your control. You set the vision early and then tighten the reins later. As the piece takes shape, then you can start requesting specific fixes. Often I'll ask it to delete sections that don't help, or clean up repetition when I've given it disconnected information without a clear outline. But watch carefully—sometimes it deletes things it thinks are unimportant that you actually wanted to keep. Stay engaged, but don't get too picky until the overall structure is solid.
The paradox is that loosening control often gives you better control over the outcome. Exhaust the high-level guidance first. Let the AI understand your full vision rather than getting piecemeal corrections that might conflict with each other. Save the micro-edits for last.
Writing with AI isn't about commanding every word—it's about finding that sweet spot between guidance and freedom. You have to learn to see what results your instructions have on the piece. Much of it is learning to communicate with the AI, and this will be different for each individual. Like developing a working relationship with a colleague, you figure out through practice what approaches yield the best results for your particular style and needs.
Think like a director, not a puppeteer—and you'll get a better show.