The gap in most AI advice is the last inch: you're sold on "AI can do this," and then you're alone with an empty chat box. This post is the missing piece — a library of prompts you can steal verbatim, organized by the work they do, each one shaped by the same anatomy we've used all series. Swap the bracketed parts for your business. The review stays yours.
The four-part anatomy (from Ask, Don't Click): every good ask names where (the page, the form, the record), what (the change or output you want), why (the goal — it changes the choices the AI makes), and what stays (the thing it must not touch). It isn't syntax; it's completeness — write it the way you'd brief a capable assistant on their first week.
How to use this library
Three rules govern every card. First, steal, then specialize: take the prompt as written, replace the brackets, and after it works three times, promote it to a workspace skill so it stops being typing at all. Second, these assume your voice guide is installed — prompts choose the work; the guide chooses the words. Third, the review gate applies to every output here, scaled to stakes: a glance for the social caption, the full checklist for the proposal.
Website
The second one is the sleeper: "every place that mentions" is exactly the sweep humans do badly. Deep dive: Ask, Don't Click — and the reason these are safe at all is your design system.
Forms and leads
The first builds both halves (form + page); the second is the journey, created because you asked — the whole wiring story is in Forms That Feed Themselves.
Email and campaigns
Theme-driven so the monthly ask is fill-in-the-blank — that's the annual plan paying rent. Which send type for which job: the decision guide.
Social
The month-in-one-sitting mechanics: the briefing workflow; carousel craft: posts that stop thumbs.
Customers and records
The first is the two-minute brief that makes you the person who remembers everything; the second is the monthly hygiene scan — note its last clause, which is the whole division of labor.
Money and the week
The first is the weekly digest opener — worth a quick action. The second's closing phrase is doing real work: numbers-not-verdicts keeps the assistant a reporter, with judgment staying yours.
Meetings and documents
"Leave pricing blank for me" is the four-part anatomy's what stays doing its finest work — pricing is never delegated. The chains: meeting to follow-up and proposals with the gate.
Reviews and reputation
The parenthetical "(ask me)" is the guardrail: the public promise of a fix is never the draft's to invent. Full treatment: replies without the robot voice.
Make the good ones permanent
A prompt you've typed three times is a skill you haven't created yet. Promote the keepers — the digest opener, the meeting brief, the newsletter format — into workspace skills with your specifics baked in, and the library stops being something you consult and becomes how your workspace simply works. Quick actions for the weekly rhythms, skills for the formats, this page for everything you haven't promoted yet.
Key takeaways
- The four parts are completeness, not syntax: where, what, why, what stays — brief the AI like a capable assistant on their first week.
- "What stays" is the safety rail: "don't change anything else," "leave pricing blank," "lists only — I'll decide" are where these prompts earn their keep.
- Steal, then specialize, then promote: verbatim first, brackets swapped, and anything typed three times becomes a skill with your specifics baked in.
- Prompts choose the work; the voice guide chooses the words: the library assumes your voice is installed — that's why none of these specify tone.
- Numbers, not verdicts: asks that end in "names and numbers" keep the AI a reporter and you the judge — the pattern behind the digest, the hygiene scan, and the overdue list.
- The gate scales with stakes: a glance for the caption, the checklist for the proposal, your own keyboard for anything requiring courage.
Frequently asked questions
Why do these prompts work better inside my workspace than in a generic chatbot?
Because half of every prompt here is a reference the AI has to be able to resolve: "the Hendersons' record," "the services page," "this month's theme," "our winning format." A generic chatbot gets eloquent guesses; a workspace AI gets your actual data, which is the difference between a draft you rewrite and a draft you review. The prompts are portable — the context isn't.
What makes a prompt fail, most often?
A missing fourth part. Asks that name where, what, and why but skip "what stays" hand the AI more canvas than you meant to — and it will helpfully repaint it. The second most common failure is bundling: three unrelated asks in one message, which gets you three half-jobs. One task per ask, edges stated, and most "AI got it wrong" stories become "AI did what I accidentally asked."
Should I save my own variants somewhere?
Yes — but as skills and quick actions, not as a document of prompts. A prompt file is the design-system-in-a-PDF problem all over again: it gathers dust while everyone retypes from memory. The working version lives where the work happens: rhythms as one-click actions, formats as skills, and only the genuinely occasional asks left as muscle memory. Your library should shrink as your workspace learns.
How do I write a good prompt for something not on this list?
Run the anatomy as questions: where does this work live? What exactly should exist when it's done? Why — what's it for? And what must not change? Then add the guardrail that matches the stakes — "show me before publishing," "ask me before deciding," "numbers only." If you can answer those five things, you can brief the work; if you can't, the prompt isn't the problem — the task isn't defined yet, and that's worth knowing before anyone, human or AI, starts.
Ready to stop staring at the empty chat box? Faster gives these prompts something to resolve against — your pages, records, money, and meetings in one workspace, with skills and quick actions for the keepers. Start free and steal the digest opener first.