The Best Way to Write Articles With AI Is to Let It Interview You

I'm writing one new thing every day for 100 days straight, weekends included. I'm on day six.

If you'd asked me a week ago how I planned to keep that pace, I'd have given you the obvious answer. Let AI write the posts. Open a chat, type "write me an article about X," clean it up, hit publish.

I tried it. It's fast. It also produced the most forgettable writing I've ever put my name on.

My first post was about using Claude to turn a call transcript into a case study. Useful topic. The draft came back clean and organized and fine. Then I read it again and thought, anyone could have written this. They'd prompt the AI the same way and get the same thing. Good information, zero me. It was the polite kind of AI slop, the kind that's good enough that you almost publish it.

The next day I changed one thing. Before I wrote a word, I told the AI: I need to be in this one. So instead of asking it to write, I had it ask me.

That post was about learning GitHub as a marketer. Why I ended up there, what I made of it, why it mattered. Same tool, same week, a completely different piece. It had a point of view. It had me in it.

Here's the thing nobody says out loud. If AI writes your article, a human still has to edit it. So you're in the process either way. The only real question is where you stand in it.

Most people stand at the back. They let AI write, then they fix the prose. That's the worst seat in the house. You burn your energy editing sentences, and the part that actually makes writing worth reading, the lived experience, the opinion, the story, never makes it in. The AI didn't have those things. You did. And nobody asked you for them.

So move to the front. Be the source, not the editor.

This one took me a minute to admit. AI is a better writer than I am. Most days it's a better writer than you too, and that's fine. Writing was never the rare part. The rare part is the thing you lived. AI can't invent what you actually went through, the tool you had to learn that nobody expected a marketer to touch, the opinion you've earned and still can't fully defend. But the second you say those things out loud, AI is great at shaping them into something people want to read.

Here's the whole thing.

Pick your subject. Tell the AI to interview you about it, like a journalist who's going to write the piece. One question at a time. You answer, and its next question follows your last answer, so it keeps digging where the good material is.

That pacing matters more than it sounds. When the AI asked me about GitHub, it didn't stop at "what happened." It asked why I went there, what my first experience was like, what I actually thought. Those questions pulled out the parts that were really me. I'd never have typed them into a prompt. I didn't even know they were the story until someone asked.

When it has what it needs, you let it write. Now it's working from your raw material instead of the internet's average, and the draft comes back sounding like you. Because it is you, just better organized.

There's a speed problem hiding in all of this. Answering good questions by typing is slow, and I've got a calendar full of meetings. Ten days in, I'd have quit.

So I stopped typing my answers and started talking them. I use Wispr Flow. You hold a button, you talk, and it turns your voice into clean, formatted text with almost no mistakes.

I tried the built-in microphone in the Claude app first. It was rough. It picked up other voices in the room, it formatted things wrong, and it cut me off. The second I paused to think, it jumped in and ran. Wispr Flow just records. I push the button, and I can sit and think for a full minute before I say a word. It waits.

That sounds small. It isn't. The thinking time is where the story lives. A tool that punishes you for pausing punishes the exact thing you're there to do. Wispr Flow has a free plan, so it costs nothing to find out if it works for you.

Here's the part most people skip, and it's the part I care about as much as the writing.

A great article nobody can find might as well not exist. So when the draft is done, I don't paste it into a CMS and walk away. I build it into the site properly with Claude Code. Clean metadata so it shows up right in search. A real title and description. Internal links to the other things I've written, so the piece isn't an island. Structured data so AI search engines can read it and cite it. The boring machinery that decides whether anyone ever sees what you made. The format of what you write matters too: listicles get cited by AI search at a rate that dwarfs any other format, and there are seven specific rules that separate the ones AI quotes from the ones it ignores.

I'm a marketer, not an engineer. A year ago that sentence would have ended the conversation. It doesn't anymore, and that's the real lesson buried in my GitHub story. Marketers are becoming builders, and AI is what unlocked it. We're not just content creators or campaign managers. We build the systems that get the results. That's the gap between a 10x marketer and one who does the same thing every day. The repeatable parts of this workflow live as Claude skills, so the AI starts every job from my best work instead of a blank page.

This post went through that exact process. I picked the subject. I had the AI interview me, one question at a time, while I talked my answers into Wispr Flow. Then it wrote, and we built it into this site together. You're standing inside the example.

I have one test for all of it. Before I publish, I ask myself: would I read this if someone else wrote it?

For the case-studies post, the honest answer was no. It was fine. Fine doesn't get read. And if I wouldn't read it, I can't expect anyone else to.

The interview version passed, because there was a real person in it. That's the whole game.

So go try it. Pick something you actually know. Have an AI interview you about it, talk your answers instead of typing them, and see if what comes out sounds more like you than anything you've prompted before. Then build it somewhere people can find it.

We're early in all of this. There's no blueprint for writing with AI yet, and the closest thing we have to one is getting rewritten the next day. The only way to find what works for you is to try it before you're ready, today, on something small. So go make something that sounds like you.