AI Wrote My First Article Draft. Then I Spent 6 Hours Fixing It.
The tools, the workflow, and what still requires a human to write a history newsletter
Past Passport is a history newsletter first. But I’m an open book about how I use AI for drafting and research. Today’s AI-focused newsletter covers what I’ve learned this week and the mistakes I made. AI not your thing? Skip it. Next history newsletter hits your inbox Friday.
Many successful writers were doing well before AI came along. Now they’re using it to get even better. That grabbed my attention last month in a three-week content creator class.
If creators earning six figures from their newsletters use AI, I think others should take notice. I did.
Now I use AI to handle certain parts of the process of creating my history newsletter. I’m open about it.
Every History Newsletter Starts with Grok
When I get an idea for a piece, I do the initial research in Grok. From what I can tell, it’s currently the strongest tool for deep historical research. It pulls specific, well-sourced material with fewer hallucinations than the alternatives. I’m not a historian with a Ph.D. in Ancient History. Grok gets me to good raw material faster than I could find it on my own.
Another option is to use an AI prompt to search for news stories in your niche. I’m still experimenting with that process. Looks promising.
Once I have enough to work with, I do a voice dump. I talk through my own version of the piece: what I find interesting, what angle I’m taking, what I want to say. That part is mine.
Then I Drop the Research Into Claude
I bring the Grok research, and my voice dump of the rough draft into Claude which handles the long-form drafting. People in my writing community kept pointing me toward Claude for this, and after trying several tools, I think they’re right. Claude handles the writing of longer essays better than anything else I’ve used.
But the first draft still doesn’t sound like me, and it’s rough. But it saves me hours of creating a bad first draft.
Then I pull it into Google Docs and start cutting.
Editing Is Where I Put On My Work Hat
I run the draft through a set of editorial prompts that review the content from different angles, each one looking for something different. It’s like having a good editor available at 2:46 in the morning that never needs a Monster to stay alert.
A human editor would often do better. But a human editor costs real money. A single pass on several newsletters weekly would cost hundreds of dollars. For a small creator working alone, that’s not realistic. AI with the right prompts is a good first-pass editor. Not perfect. Better than nothing, and a lot cheaper than the alternative.
And I’m learning from them. I’m starting to anticipate the notes before they come.
The most common note I get is: “You’ve written two essays and tried to publish them as one.”
I’ll have a pile of good research, and I want to use all of it, so I keep adding, and the point gets buried. A 1,000-word essay can’t go in three directions. Cutting solid research feels wasteful every time. But a focused piece that persuades beats a thorough piece that doesn’t.
After the editorial pass, I run an anti-AI-slop prompt designed to strip out the obvious tells. Then I edit by hand.
Today’s draft had about a dozen em dashes. I have a prompt that explicitly bans em dashes. The AI put them back in anyway. It also reinserted “actually” and “genuinely,” two words I’ve told it, in plain language, are not allowed. They showed up anyway. This is how it works. The model was trained on patterns, and some of those patterns don’t respond well to instructions.
What I’ve Learned After 30 Days
My eye for AI writing is sharper than it was. I can feel when a sentence isn’t mine, even when I can’t explain why.
I think AI is saving me time overall. But I’m doing 6 to 8 editing passes before I publish. It’s different work. The output is better, but every essay still takes me several hours.
I know creators who make serious money and use AI for research, drafting, and editing. They’ve figured out how to use it without losing their voice.
Does it still sound like you when it’s done? If yes, the process that got you there doesn’t matter.
If you’re just starting out with AI, here’s what I’d recommend:
Start talking, stop typing. Do a voice dump of your first draft instead of typing it. I use Wispr Flow, it’s free, and it changed how I work. Took a few days to get used to. Now I can’t imagine going back.
Build a voice profile. This is the single most important thing. Claude needs to know how you write before it can write like you. A writer named Ruben Hassid runs a newsletter called How to AI. He gives away the exact prompt used to build your voice file for free. Start there: ruben.substack.com/p/youre-just-a-text-file
Use Grok for research, Claude for writing. Grok pulls deep historical material fast. Claude handles long-form structure better than anything else I’ve tried. Don’t use the same tool for both jobs.
Plan to edit more than you write. The first draft is a rough starting point, not a finish line. I run 6 to 8 passes before I publish.
Find an anti-slop prompt and run it. There are prompts out there designed to strip AI tells from a draft. I can’t share the ones I use because I got them from other writers, but search for them. They exist, and they help. Run one before you think you’re done. Ryan Stax publishes The AI Newsletter, which has good anti-AI slop prompts.
Then edit it yourself anyway. Even after the anti-slop pass, AI will reinsert junk. Em dashes come back. “Actually” and “genuinely” show up again. The model was trained on patterns it can’t fully unlearn. A final read by your human eyes is essential.
An optimist writing about history and using AI to create his newsletter. Subscribe for three history essays per week and regular updates about AI for non-technical creators.
This is the most honest take on AI writing. It can save you time, but it does not save you from having taste.
And never let the AI replace your voice.
Great article!