I used to write a fresh, careful prompt every single time I opened an AI chat. Like a fool.
Then I noticed something embarrassing: I was basically rewriting the same seven prompts over and over, just with different nouns. So I saved them. Now I paste, fill in the blank, and move on. My output got better and faster, which almost never happens together.
Here's the whole library. Steal it. They're not clever. They're just the ones that earn their keep every week.
Most people get mediocre results from AI because they ask vague, one-line questions. The fix isn't longer prompts — it's reusable prompts with structure: a role, a goal, constraints, and a requested format. Below are the seven I actually reuse, covering writing, planning, research, decisions, learning, critique, and editing. Copy them, swap the brackets, and you'll never stare at a blank chat box again.
The dirty secret of prompting is that consistency beats cleverness. A decent prompt you reuse 200 times will out-deliver a brilliant prompt you write once and forget.
A reusable prompt also forces you to think clearly about what you want before you ask. Half the value isn't the AI — it's that filling in the brackets makes you specify the goal you were being lazy about. Most bad AI output is really a bad question wearing a costume. That's the thread running through the honest truth about AI productivity tools: the leverage is in how you ask, not in which model you pay for.
It's also why the advice to stop using ChatGPT like a search engine matters — a saved template nudges you toward giving the model material to work with instead of firing off a one-line lookup.
So I treat prompts like saved code snippets. Tested. Named. Reused. Here they are.
I spent years saving the hardest task for when I 'felt ready.' Doing it first instead quietly fixed my focus, my dread, and my output.
I tracked every distraction for a week and was horrified by what I found. Then I fixed the three that mattered most.
I went from 200 to 11,000 subscribers without hiring anyone. AI didn't write my newsletter — it did everything around it.
This is the workhorse. I use it for emails, posts, intros, anything.
Write [thing] for [audience]. The goal is [specific outcome]. Voice: [direct / warm / blunt]. Keep it under [N] words. Give me 3 versions that differ in opening line, not just wording.
The magic line is "3 versions that differ in opening line." It stops the AI from giving you three near-identical drafts and forces real variety where it matters most — the hook.
When a task feels foggy, I make AI turn it into steps before I touch it.
I need to [goal] by [deadline]. Break this into the smallest sequence of concrete steps. Flag the one step most likely to cause delay, and tell me what I could do today to de-risk it.
That last clause — "what I could do today" — is what makes a plan usable instead of decorative. A plan I can't start is just a nicer-looking version of dread.
For getting up to speed on something unfamiliar fast.
Explain [topic] to me as if I'm smart but completely new to it. Start with the one core idea everything else hangs off. Then give me the 5 things I most need to know, and 2 common misconceptions. End with what an expert disagrees about.
The "what experts disagree about" line is gold. It tells you where the real edges are, instead of a flat Wikipedia summary that pretends everything is settled.
I reach for this whenever I'm stuck between options.
I'm deciding between [option A] and [option B]. My priorities, in order, are: [1], [2], [3]. Argue hard for each option as if you believed it. Then tell me which you'd pick given my priorities, and the single biggest risk I'm underweighting.
Making it argue both sides before recommending one is what keeps it honest. A single-sided AI answer just flatters whatever you already leaned toward.
When I want to actually retain something, not just read it.
Teach me [concept]. After your explanation, quiz me with 3 questions of increasing difficulty. Wait for my answers before telling me if I'm right. If I'm wrong, explain why simply, then ask again.
This turns a passive read into a tiny tutoring session. The "wait for my answers" instruction is doing heavy lifting — it stops the AI from helpfully spoiling the quiz it just wrote.
For anything I made and secretly suspect is flawed.
Here is [my draft / plan / idea]. Don't be nice. Find the 3 weakest points and the single thing most likely to make this fail. For each, suggest a specific fix. Assume I can handle blunt feedback.
"Don't be nice" genuinely changes the output. AI defaults to encouraging — a tendency Harvard Business Review has flagged in its coverage of how AI assistants tend to flatter rather than challenge their users. You have to give it explicit permission to tell you the thing you don't want to hear.
The final pass on any writing.
Tighten this without changing my voice. Cut filler, kill weak verbs, and flag any sentence that's trying too hard. Show me the edited version, then list the 3 biggest changes you made and why.
The "list the changes" part means you learn from the edit instead of just accepting it. After a few months, I started catching my own filler before the AI did.
There's one move that doubles the value of every prompt above, and almost nobody does it: chaining.
A single prompt produces a single artifact. But real work is rarely one step. So I run them in sequence, feeding each output into the next. The planning prompt produces a plan. I pass that plan straight into the critique prompt — "find the weakest step." I take the fixed plan into the writing prompt to draft the first deliverable. Then the editing prompt for the final pass.
Four prompts, one continuous flow, and each one operates on the real output of the last instead of a fresh blank context. The quality compounds. By the end, the AI has accumulated context across the whole task, and the final result is sharper than any single mega-prompt could produce.
Here's a chain I run constantly for any new piece of work:
Five saved templates, zero original thinking about how to ask, and a finished thing at the end. The prompts aren't just shortcuts anymore — chained together, they're a tiny assembly line.
The other thing chaining taught me: keep each prompt narrow. A prompt that tries to plan and write and edit in one go produces mush. Five focused prompts run in sequence beat one ambitious prompt every time. Narrow tools, chained, win.
I keep all seven in a single note pinned to my desktop. When I need one, I copy it, fill the brackets, paste. Ten seconds.
Once a month or so I tune a phrase that's drifting. That's the entire maintenance cost. A few of these have lived in my library for over a year now, which in AI terms is basically forever.
You can build the same thing in any notes app, or in whatever AI assistants you already use. The container doesn't matter. Having the templates ready is the whole win.
One small upgrade that paid off: I gave each prompt a one-word nickname. "Sharpen" is the critique prompt. "Sketch" is the writing one. "Map" is planning. Now I don't even paste the full text for the ones I use most — I've got them memorized by feel, and the nickname is just how I think about which tool I'm reaching for. It sounds trivial, but naming a thing is how it stops being a document and starts being a habit. The library went from something I consulted to something I simply use, the way you don't think about reaching for a specific wrench.
I'd also gently warn against hoarding. I tried, at one point, to build a library of forty prompts — one for every conceivable situation. It collapsed under its own weight. I never remembered which was which, and choosing between forty options is its own kind of friction. Seven is the number that stuck precisely because it's small enough to hold in my head. If you find yourself with twenty prompts, you probably have five good ones buried in fifteen you'll never reach for. Cut to the ones you actually reuse, and let the rest go.
If even one of these saves you a blank-chat-box morning this week, build your own version of the list and keep refining it — that's the whole game.
Q: Won't these stop working as AI models change? The structure outlives the models. Role, goal, constraints, format — that scaffold has held up across every model I've used. I tweak wording occasionally; I've never had to rewrite one from scratch.
Q: Should prompts be long or short? As long as they need to be to remove ambiguity, and not one word longer. My longest is six lines. If a prompt feels long, it's usually because the task was vague and you're finally being specific.
Q: Do I really need all seven? Start with writing, planning, and critique. Those three cover most of what most people do. Add the rest when you feel the gap.
Q: Can I combine them? Yes. I'll often plan, then critique the plan, then write from it. Chaining your saved prompts is where the real productivity compounding happens.
You don't need to be a "prompt engineer." You need seven good templates and the discipline to reuse them.
The people getting incredible results from AI aren't writing magic spells. They're pasting the same boring, well-structured prompts they refined months ago — and spending their saved brainpower on the actual work.
Save these. Tweak them until they sound like you. Then watch how much faster you move when the blank chat box stops being a blank page.
Which of these would you reach for first tomorrow morning?
Comments
Sign in to join the conversation
No comments yet. Be the first to share your thoughts!