Reply to any email
The everyday workhorse — you know what you want to say, you want it to land well.
57 prompts · 9 categories
Copy, swap the [brackets], use. Every prompt here is a working template for something people actually do — replying to email, researching a decision, writing a post that isn’t slop, generating an image that matches the one in your head.
Tool tags mark where a prompt is known to work well. Any means it is tool-agnostic — the technique does not depend on the vendor. If you want to understand why these are shaped the way they are, read the course.
The everyday workhorse — you know what you want to say, you want it to land well.
They went quiet. You need an answer without sounding like a debt collector.
Declining a request, invitation, or piece of work you cannot take on.
A slip, a mistake, a price rise, a cancellation. The email you are dreading.
You got CC'd into 40 messages and need to know if you matter.
Run this on anything written while angry, or anything going to someone senior.
First contact with someone who has no reason to care.
Firm, professional, doesn't torch the client relationship.
You need to understand a topic properly and be able to check the answer.
Someone asserted something confidently. You want to know if it holds.
Tools, vendors, approaches — anything where you need to pick one.
A long report, contract, or spec that someone needs to act on.
Deeper research where you need the state of the evidence, not a blog summary.
Big research task where you are not sure what you are actually asking.
A real post — one with a point, built from things you actually know.
A post that cites real figures — and doesn't invent them.
Your draft exists. You need to know what's wrong with it, not have it replaced.
Too long, and every sentence feels necessary. They aren't.
You have the material. The ordering is the hard part.
Set this once per project. Everything after it sounds like you.
Recurring format, needs to not feel like a template every week.
The default LinkedIn voice is a parody. This bans it explicitly.
Something happened at work. There's a post in it, but not the obvious one.
Add something to someone's post instead of 'Great insight!'
Your profile currently says “Passionate about driving innovation”.
New job, launch, milestone. Needs to be said without “I'm thrilled to share”.
Raw notes or a transcript, turned into what actually got decided.
You're stuck between options and going in circles.
Planning that accounts for the fact that your week will not go to plan.
Learning a new concept, tool, or domain from zero.
Before you commit to something you're excited about.
Receipts, listings, notes, form responses — anything with a shape.
Excel or Sheets formula you can actually verify.
The order that works: subject, action, setting, style, light, framing, quality.
When it needs to look like a photograph, not an illustration.
Almost right. One thing keeps going wrong.
You have an image you like. You want the prompt that makes more like it.
Blog headers, social cards, product illustrations that match a look.
Same person or mascot across multiple images.
Hero, subhead, sections — for a product that does something specific.
You know what you want. It needs to be written down before anyone builds it.
A second pair of eyes that isn't agreeable.
Something broke. Give the model evidence, not your theory.
A brief that produces a page worth ranking, not keyword mush.
You know the outcome you want. You don't want to engineer the prompt.
You wrote a prompt. The output is wrong and you don't know why.
Anything where you need to check the answer, not just receive it.
You have a conclusion. You want to know if it survives someone who doesn't own it.
Set once in Claude Projects or ChatGPT custom instructions. Applies to everything.
The task is too big for one prompt and you're getting mush.
You gave six requirements and got output covering four.
The text is fine but reads as generated. This names the specific tells.
Learn what your generated text keeps doing wrong, so you stop shipping it.
Generated draft, your voice. Samples do what adjectives cannot.
The text is vague. Vagueness is the deepest AI tell, and the hardest to fix.
Text that's technically fine but nobody would ever say out loud.
Twenty turns in, everything you say is 'a great point'. Time to reset.