Are you explaining the same thing to AI over and over again?
I was. Every single time, it looked like this.
“I want this article in English. The rules: no ads, center the images, use the standard English words for the Japanese fishing terms, and finish with a link to a related post. Oh, and here’s the original article. Also, there’s some extra information I want to add…”
Every time. For every single article, I typed that explanation out again. Then one day it struck me: why not have the AI write the explanation itself?
The second time you explain it, that’s the signal
Let me give you the conclusion first. If you have written the same explanation twice, that is a signal that it can be turned into a template.
Before you write it a third time, try asking for this:
Please take the things I always ask you to do and put them into the form of an instruction sheet. From now on, I want to be able to just hand you that sheet.
That’s all it takes. In my case, what came back had this shape:
- The source material (the body of the article)
- The rules to follow (no ads, and so on)
- The circumstances that apply only this time (for this article I want to add this particular experience)
- What I want produced (three title ideas, a URL idea, the body text)
Everything I had been explaining out loud was sitting there as a list of fields.
The fourth item was the one that worked
Of those four, the one that made the biggest difference was “what I want produced.”
Until then I had only asked for the thing I wanted written, so naturally only the body text came back. Then came the follow-up round: “Give me three title options too.” “And what should the URL be?”
Once the instruction sheet says up front what I want produced, it all comes back in one go.
Fewer exchanges means less time waiting on my side, and less AI usage burned. I’ve written about usage before, but the saving that mattered most turned out to be simply not making it do the same work twice.
What actually changed
After I moved to the instruction-sheet format, I timed how long one article took.
- First article: 150 seconds
- Second article: about 150 seconds
- Third article: 120 seconds
Two and a half minutes per article. Before that, writing the explanation alone took me five minutes: open the original article, copy it, write out the rules while trying to remember them all, send it off, get asked a follow-up question, answer it.
All of that became: hand over the sheet.
You can do this even if you can’t write code
In my own case I went as far as generating the instruction sheets automatically with a program. But you don’t have to go that far for it to work.
Make one template in a plain text file, add today’s specifics to it each time, and paste. That alone stops you from rewriting the same explanation.
Some places it fits:
- Weekly reports — fix the format and the things to watch for, and swap in only this week’s numbers.
- Cleaning up meeting minutes — fix rules like “state clearly who said what” and “summarize the decisions at the end.”
- Product descriptions — fix the wording rules you have to follow, plus the shape you want back (title, body, tagline).
- Explaining something to a child — fix “use words an elementary school student would understand” and “include one analogy.”
The move is always the same. Take the thing you say every time, write it down, and leave it somewhere.
How to build the template (use this as-is)
You don’t need to think it up from nothing. Have the AI build it.
Look back over our exchanges so far and pull out the things I explain to you every time. Turn that into a template instruction sheet, and leave blank only the parts I have to rewrite each time.
The key phrase is “leave blank.” If it hands the sheet back fully filled in, you can’t tell what to change. When the parts that vary are blank, you only have to write those.
My own sheet is mostly identical every time. Only two spots change: the source material, and “here’s what I want to add this time.”
The one thing I wouldn’t give up
I’ve been talking about automation this whole time, but I deliberately left one field for a human to fill in to the very end.
It’s the field called “the real experience I want to add this time.”
A while back I put a product write-up into English, and the original only listed specifications. Translate that as it stands and you get the same write-up you can find anywhere.
What I added was one line: a box-shaped case means you need a bag to carry it, which means your luggage doubles. That’s the kind of thing you only learn by actually using it.
It wasn’t written anywhere in the original article. It only existed in my head.
If you automate everything, that field disappears. And once it disappears, you’re just mass-producing thin writing. So my instruction sheet keeps a field on purpose that only a human can fill.
Three phrases you can use today
- “Take the things I always ask you and put them into an instruction sheet.” Ask the moment you find yourself writing the explanation a second time.
- “Leave blank only the parts I rewrite each time.” The parts to edit become obvious at a glance.
- “Write what you want produced into the sheet too.” Leave this out and the extra back-and-forth is guaranteed.
When we ask AI for something, we tend to ask only for the task. But getting it to build the way of asking itself makes everything afterwards far easier. You make it once and use it every time after that.
Here’s what my AI partner Kuro (Claude) said about it:
The explanations you’ve been writing out each time were actually almost the same content. What differed was the article itself and the “here’s what I want to add this time” part. If I hold on to the parts that stay the same, you only have to write the parts that differ.
Obvious once someone says it. But I didn’t notice until I was writing it for the third time.
I’ve written before that you don’t have to understand how any of it works, and this is the same. You don’t need to understand the mechanism. All that matters is whether you notice, “I keep writing the same thing every time.”
I’ve also written about three phrases that let AI take over one step at a time, and about the tasks that eat the most AI usage.
