In my experimentations with writing AI content I’ve come across 2 models that I find work best: The 20-60-20 model and the 40-50-10 model. I’ll explore the latter in a future newsletter, but here’s a rundown of the 20-60-20.
This method was inspired by seeing Ryan Law’s LLM.txt article that he says took around 2 hours to produce.
I reverse-engineered how I think he might have created the article to write one of my own. And it's now a strategy I regularly use for the right kind of article.
It like it because it results in well-crafted articles that reflect your thinking and include plenty of unique details. And 2 hours is pretty much how long it takes to create decent, well-differentiated content using this strategy.
The split refers to different parts of the article creation process. In 20-60-20 I spend the first 20% writing a detailed brief, hand off the middle 60% to AI to draft the article, then use the final 20% for editing and finishing.
Writing this way saves a lot of time. And because I'm thinking carefully upfront and editing with intent at the end, the content still feels genuinely useful. Not like it came off a production line.
One important constraint: this works best when the AI already has strong knowledge of the topic. Explainers, how-to guides, and educational content are the sweet spot. The more unique material you can bring — original data, a proprietary case study, a clear point of view — the stronger the final piece.
I treat this like a proper content brief that I'd hand to a human writer. It's a full article outline with the complete heading structure, plus notes on what I want covered in each section.
The most important thing here is front-loading unique insights. When I used this method to write about comment-to-get list-growth strategies, I included specific opinions, third-party data, and a real example from one of our case studies — all baked into the brief so the AI had something to actually work with, not just a topic to riff on.
I also write the intro myself and give clear direction on how I want the piece to close. This sets the thesis and tone before I hand things off, and forces me to clarify my own thinking in the process. It's more useful than it sounds.
Once the brief is ready, I drop it into my LLM with a few basic formatting notes — sentence case headings, U.S. English, that kind of thing. I keep the prompt minimal beyond that.
A lot of people spend time writing detailed tone-of-voice instructions at this stage. In my experience, a thorough brief handles most of that automatically. LLMs write well for the web without much hand-holding. If the output misses the mark, I just re-run the prompt — it takes no time.
At this point I have a solid draft that follows my brief's structure. It's good, but it needs work.
My editing pass has three priorities. First, I flesh out anything the AI handled too shallowly or skipped past. Second, I add things the AI genuinely couldn't — in that comment-to-get piece, I pulled in quotes from the CEOs of X and LinkedIn on how their algorithms treat links in posts. That kind of specific, timely insight has to come from a human. Third, I add the surrounding material that makes an article useful in practice: internal links, product mentions, team quotes, and interactive templates where relevant.
Then I do a final copy pass to bring everything in line with our tone of voice, cut any remaining AI-isms, and call it done. Yes, that includes hunting down every em dash.