How I Turn My Best Prompts Into Reusable Templates
For about a year, every good prompt I wrote died the moment I closed the tab. I'd spend twenty minutes crafting a prompt that finally got Claude to write product copy the way I wanted — the right banlist, the right structure, the right voice notes — use it once, and then a week later, faced with the same task, start from a blank box. I wasn't reusing my best work. I was re-deriving it, badly, every single time.
The tell was that my output quality was a sawtooth. Some weeks the copy was sharp because I happened to reconstruct the good prompt. Some weeks it was generic because I rushed and wrote a lazy one. The model wasn't inconsistent. My prompts were.
The fix was embarrassingly simple once I saw it: a great prompt isn't a one-time message, it's a template with holes in it. The work you did to make it good is reusable. The only part that changes between uses is the specific subject. Separate those two things — the reusable scaffold and the swappable slots — and you never rewrite a good prompt again.
Here's the system I use, and the real templates I run every week.
Why rewriting from scratch is a hidden tax
When you rewrite a prompt from memory each time, three things go wrong, and none of them are obvious in the moment.
You lose the hard-won parts first. The banlist, the "what does this sacrifice" question, the specific structural constraint — those are the parts that took experimentation to land on, and they're exactly the parts you forget when you're typing fast. You keep the easy instructions and drop the ones that did the real work.
You can't improve a prompt you don't keep. If every prompt is disposable, you never iterate. A template you reuse fifteen times gets sharpened fifteen times. A prompt you write once and throw away is frozen at version one forever.
Your quality floor is set by your laziest day. Without a saved template, the prompt you use is whatever you can be bothered to type right now. On a good day that's great. On a busy day it's "write me a launch tweet," and you're back to sludge.
A template fixes all three. The hard-won parts are baked in permanently. Every use is a chance to refine the scaffold. And your worst day still runs your best prompt, because the good version is the one sitting in front of you.
The bracket-and-slot system
The whole method is this: take a prompt that worked, and replace every part that was *specific to that one use* with a labeled slot in brackets. Everything that stays the same — the voice, the constraints, the output format — stays as fixed text.
Take a real example. Here's a prompt I once wrote freehand to draft a homepage hero for Super Prompts:
Write the homepage hero for Super Prompts, a tool that lets creators save and rediscover AI prompts. Audience: solo founders and creators drowning in scattered prompts. Headline max 8 words, subhead max 20 words, one CTA. No "transform your workflow," no "unlock," no exclamation marks. Direct and concrete. Read it back as plain text before formatting.
That worked. But almost none of it is actually about Super Prompts — most of it is about *how I want hero copy written*. So I turn it into a template by socketing out the three things that change:
Write the homepage hero for [PRODUCT — one-line description]. Audience: [WHO IT'S FOR + their core frustration]. Headline max 8 words, subhead max 20 words, one CTA. No "transform your workflow," no "unlock," no exclamation marks. Direct and concrete. Read it back as plain text before formatting.
Now it's reusable. Next time I write a hero — for a landing page, a side project, a client — I fill the two brackets and the entire apparatus of constraints comes along for free. The twenty minutes I spent getting the voice right is permanently banked.
The rule for what becomes a slot: if it would be different the next time you run this, it's a slot. If it's the same every time, it's scaffold. Product name, audience, topic, tone target — those vary, so they're brackets. Banlists, structure, output format, voice — those are constant, so they're fixed text.
Template 1: The repurposing template
This is the one I run most. I write one piece of content, then template the act of cutting it into five other formats.
Here is a piece I wrote: [PASTE THE SOURCE PIECE]. Repurpose it into [FORMAT — e.g. a 200-word LinkedIn post / a 5-tweet thread / a short video script]. Keep the single strongest idea and cut everything else — don't summarize the whole thing, amputate it down to the one point that survives in this format. Match the voice of the original. No "in today's landscape," no hashtags, no emojis. Before you write, tell me in one line which idea you're keeping and which you're cutting, then write it.
Two slots — the source and the target format. The "tell me what you're keeping first" line is the part that took me a dozen tries to land on, and now it rides along every single time I repurpose anything. That's the whole point of the template: I never have to remember to add it again.
Template 2: The critique-before-I-ship template
I never publish anything important without running it through this. It's a template because the *kind* of critique I want is constant — only the artifact changes.
Here is [WHAT IT IS — a landing page / cold email / feature spec] I'm about to ship: [PASTE IT]. Don't praise it and don't rewrite it. Do three things: (1) name the single weakest sentence and why it's weak, (2) tell me the one objection a skeptical [READER TYPE — e.g. a busy founder who's seen 100 of these] would have, (3) tell me what I'm assuming the reader already knows that they might not. Be blunt — I'd rather hear it now than from the market.
The reader-type slot is what makes this portable. Same critique machinery, pointed at a different audience each time. I've run this template on copy, pricing pages, even personal emails — the scaffold doesn't change, just the two brackets.
Template 3: The decision-stress-test template
When I'm weighing a real decision, I don't want to re-derive a good thinking prompt under pressure. I want to fill two slots and go.
I'm deciding whether to [DECISION]. Context you need: [2-3 LINES OF CONSTRAINTS — money, time, what's reversible]. Give me the strongest case *against* doing it, the strongest case *for waiting and doing a smaller version*, and the single piece of evidence that would tell me I'm wrong. No hedging, no "it depends." End with: do it now, do it smaller, or don't do it.
Everything except the two brackets is the reusable thinking structure. The decision and its context change every time; the demand for a sharp, non-hedged verdict never does.
How to know a prompt is worth templating
Not every prompt earns a slot in your library. The signal is repetition with variation. I template a prompt when I notice I've now written *roughly the same thing* two or three times for different subjects. That's the moment to stop, look at the last good version, bracket out what varied between runs, and save the scaffold.
The other signal is effort. If a prompt took real experimentation to get right — multiple iterations, a banlist I tuned, a structure I fought for — it's too expensive to throw away. Anything that cost you twenty minutes to perfect should never have to be perfected twice.
If a prompt is one-off and trivial, don't template it. You're not building a museum of every message you've ever sent. You're keeping the small number of scaffolds you run again and again.
What I'd do if I were starting today
Open your AI chat history and scroll back through the last two weeks. Find one prompt that worked — one where the output was genuinely good. Look at it and ask: which parts of this were about *this specific task*, and which parts were about *how I want this kind of task done*?
Replace the first kind with `[BRACKETS]`. Leave the second kind alone. Save what's left.
That's a template. The next time the same task shows up, you'll fill two slots instead of staring at a blank box trying to remember what made it good last time. Do this for your five most-repeated tasks and you've turned your best prompts from disposable messages into compounding assets.
A great prompt you wrote once is a nice afternoon. A great prompt you templated is a tool you'll use a hundred times.