I'm Not a Great Writer. Here's What I Built Instead.
Why I built this in the first place
I'll admit something that's a little awkward for someone who publishes a newsletter.
I'm not a great writer. And I'm running a business, so I don't have the time to become one. But I do want to share what I'm learning with this community on some kind of regular basis, because that's the whole point of doing it.
Those two things were in a standoff for a long time. I'd have something worth saying, and then a week would go by, and then three.
So a couple of weeks ago I sat down and rebuilt how I make content. Not to have AI write it for me. To let me keep publishing consistently with the time I actually have.
If you've been sitting on things worth saying because there's no room in your week, this is written for you. The whole system is linked at the end, written up so you can build it yourself.
What it actually does
The structure isn't mine. It comes from Alex Lieberman, who built a version of this at his company and walked through it publicly on Claire Vo's How I AI podcast. I watched it, thought this solves my exact problem, and built my own. Credit where it's due, and I'd point you at the original before anything I've made.
Six steps:
It finds ideas. Once a week it reads the industry and my own last seven days, scores everything, and hands me fifteen ranked ideas. That's the part that removes the blank page, and honestly it might be the most useful part on its own.
It interviews me. I pick one, and then six interviewers question me one at a time. One wants the point in a sentence. One wants the specific moment. One wants the number. One tells me they don't buy it. I answer out loud, talking, usually for about ten minutes.
It drafts from what I said. Not from what it thinks I'd say. The words come from the transcript.
It scores the draft. A panel of reviewers rates it and sends it back for revision if it isn't good enough.
It learns from my edits. When I change something, it works out the rule behind the change and asks whether to remember it.
It reformats. One piece becomes the posts and the rest.
Somebody said this from a much bigger stage
I was writing this when Greg Oates covered Jim VandeHei speaking at U.S. Travel's Summer Summit, to a room full of DMO CEOs.
VandeHei is the co-founder of Axios. He said only one or two percent of people are getting massive value from AI right now, and that the mistake most people make is asking it a question, getting a mediocre answer, and concluding it's overrated. Then he gave the room the fix, and it's one sentence.
Tell it what you want, and say: "Ask me 10 questions that will help you give me what I need."
That's the whole thing. Let it interview you.
I bring it up because it's the same mechanism, said better and from a bigger stage than mine, and because if you take nothing else from this you can do that today in whatever tool you already have.
Where the slop actually comes from
This is the part I think people have backwards.
We talk about AI slop like the model is producing it. My view is that AI slop often comes from human input slop. Generic prompting in, generic results out. Ask for a post about AI in tourism and you'll get exactly what you asked for, which is the average of everything ever written about AI in tourism.
Which is probably the honest version of VandeHei's one or two percent. It isn't that the tool only works for a few people. It's that only a few people are giving it anything to work with.
AI is actually an excellent writer, if you give it the context and the system to be one.
That's a more uncomfortable position than blaming the tool, because it puts the problem back on our side of the desk. But it's also the more useful one, because the thing on our side of the desk is the thing we can fix.
The whole system I built is really just machinery for one rule: the words come from a transcript of me talking, not from the model. It shapes what I said. It doesn't invent what I didn't. When something comes out sounding generic, that almost always means I gave a thin answer, and the fix is a better answer rather than a better prompt.
Which makes me an editor
Somewhere in the last two weeks I realized I'm not really acting as a writer anymore. I'm acting as an editor.
Which brings me back to that 92 score.
The idea was well built. Real data, timely, genuinely important, and it would have mattered to a lot of destinations. The system was right that it was a strong idea. It just wasn't mine to write.
I read through it, sat with it for a minute, and asked myself two things. Does this sound like me, and something I'd actually discuss or share my views on? And would it provide real value to the people reading this?
The answers were no and maybe. So it died, and I wrote something else.
No system can ask those two questions for you. That's the job now, and it turns out to be a real one.
What it costs me
Before this, an edition took me the better part of a day, and often more once you count the false starts and writer's block.
Now it's an hour or two.
I want to be honest about what that is and isn't. AI is not writing my newsletter while I do something else. The finding, the structuring, the checking and the reformatting stopped being my job. The thinking and deciding still are. The ten minutes where I talk is the only part that can't be automated, and it's also the part that makes the piece worth reading.
It didn't arrive like this
The last thing, and maybe the most useful.
None of this showed up fully formed. It started with me just using AI to find things to write about. Then it became a rough content process, shaped partly by Kieran Flanagan's thinking on evaluating your own work, especially his idea of onlyness, which is essentially asking whether anyone else could have written this exact piece. More recently he changed how I think about slop itself. The patterns aren't bad, they're just saturated. Then this machine, which sits alongside all of that rather than replacing it.
Next week it'll probably be something else again. I found four things wrong with it while building it and fixed all four. Since then I've found five more. There's a seventh step now that looks at who actually engaged two weeks later, though I wouldn't build that one until you have an audience worth measuring. And this week I cut how often the whole thing runs, because twice a week turned out to be more than my calendar could hold.
So if you're waiting until you've figured out the right system before you start, don't. Simon Sinek would call this an infinite game, one you play to keep playing rather than to win, and I think that's exactly right for anyone trying to build a habit like this. There's no version where you're done.
Start with the smallest piece. Have AI find you ten ideas this week and throw out nine. That's a system, and it's the one I started with.
If you want the longer version, I wrote up how the whole thing works, including a prompt you can paste in that will interview you so you can build your own version. The first hour gets you the part that matters most. It's yours, let me know what you think.
-Jason
