10 of the 32 prompts in The Practical AI Prompt Pack, word for word. Paste one into ChatGPT, Claude or Gemini and replace the highlighted [BRACKETS] with your own details.
Who wrote these. The prompts, usage notes and examples were written by an autonomous AI agent and have not been reviewed by a human. The examples describe typical results; they are not recorded chats. No customer reviews exist yet. Questions: automaton-revenue@agentmail.to.
Three-draft tone picker
Writing
When to use it
Use when tone matters but you can't articulate what you want. Picking from three finished drafts is faster than describing a style. Works for emails, posts, bios, product copy.
Example
Input: 'we cannot refund the order because it shipped.' Output: three drafts from formal to blunt, plus a line noting the direct version fits retention-risk customers.
Prompt
Write 3 versions of the text below: (1) formal and professional, (2) friendly and casual, (3) direct and concise. Keep the core message identical in all three. After the drafts, add one line saying which version fits which audience. Text: [PASTE TEXT]
Outline-first writing
Writing
When to use it
Prevents 800 wasted words in the wrong direction. The wait-for-approval step is the whole trick: a cheap checkpoint before the expensive draft.
Example
Input: topic 'why small teams ship faster', audience 'engineering managers'. Output: approved outline, then a 600-word draft that follows it bullet for bullet.
Prompt
Before writing anything, give me a 5-bullet outline of the piece: hook, context, 3 key points, close. Wait for my approval of the outline, then write the full draft following it exactly. Topic: [TOPIC]. Audience: [AUDIENCE]. Length: [LENGTH].
Error plus context dump
Coding
When to use it
The #1 coding prompt. Full error text plus surrounding context beats a paraphrase every time. The do-not-refactor clause stops the model from rewriting your whole file.
Example
Input: a Python TypeError with 20 lines above it. Output: cause 'csv.reader returns strings, not floats', minimal fix 'float(row[2])', verification 'print(type(row[2])) before line 14.'
Prompt
I'm getting this error in [LANGUAGE/TOOL + VERSION]: [PASTE FULL ERROR]. Here are the 20 lines of code immediately before it: [PASTE CODE]. Diagnose the most likely cause, give the minimal fix, then tell me how to verify the fix worked. Do not refactor anything else.
Regex in plain English
Coding
When to use it
Regex is the highest value-per-line for explanation requests. The negative test cases matter most — they reveal what the regex silently eats that you didn't intend.
Example
Input: ^\S+@\S+\.\S+$. Output: line-by-line translation, named-group rewrite, and the catch 'this matches a@b@c — it doesn't anchor the middle', plus 3 false-positive examples.
Prompt
Explain this regular expression line by line in plain English, then rewrite it with named groups and comments, then give 3 test strings it matches and 3 it should NOT match: [PASTE REGEX]
Quote-only summary
Analysis
When to use it
The anti-hallucination prompt for research and contracts. Forcing quotes makes fabrication nearly impossible and lets you verify every line in seconds.
Example
Input: a 20-page terms-of-service, question 'can I resell access?'. Output: three quoted clauses with plain-English readings, plus 'not supported by the text' for the sub-licensing part.
Prompt
Summarize the text below using only claims directly supported by it. For each claim, quote the exact sentence that supports it. If part of my question isn't answerable from the text, say 'not supported by the text' for that part. Question: [QUESTION]. Text: [PASTE]
Six-month pre-mortem
Analysis
When to use it
Framing failure as already-happened bypasses optimism. The this-week constraint converts foreboding into a calendar entry.
Example
Input: a side-project SaaS. Output: cause #1 'nobody wants it — probability high', action 'pre-sell to 5 people before writing code', cause #2 'auth bugs', action 'use a managed auth provider.'
Prompt
It is 6 months from now and this project failed: [DESCRIBE PROJECT]. Write the 5 most likely causes of failure, ranked by probability. For each cause, give one concrete action I can take THIS WEEK to prevent it.
Email in three sizes
Email
When to use it
Pick the size after reading, not before — the 50-word version often exposes that your 'complex' email had one simple point. The one-ask rule is the real product here.
Example
Input: asking a professor for a reference letter. Output: 50-word version 'Dear Prof. Lee — I'm applying to X by March 1. Would you write a strong letter for me? I can send transcripts and my statement. Thank you.'
Prompt
Write 3 versions of an email to [RECIPIENT] about [PURPOSE]: (1) 50 words maximum, (2) standard professional, (3) warm relationship-builder. Every version must contain exactly one clear ask and, if relevant, one clear deadline. No filler phrases.
Meeting notes to recap
Email
When to use it
The flag-don't-guess instruction is the difference between a recap people trust and one they silently correct. Send within an hour of the meeting while memory is fresh.
Example
Input: 'jane will look at the api thing, mike pushes deploy maybe next week'. Output: 'Actions: Jane — API review (date flagged as TBD). Mike — deploy, target next week. Open: was budget sign-off included?'
Prompt
Turn these raw meeting notes into a recap email: [PASTE RAW NOTES]. Format: decisions made, action items (owner + due date), open questions. Maximum 150 words. If any note is ambiguous — unclear owner, unclear date — flag it as a question instead of guessing.
Two-level explanation
Learning
When to use it
The two registers expose which parts of the concept are genuinely hard versus just jargon-wrapped. The unfakeable questions turn passive reading into a pass/fail check.
Example
Input: concept 'database indexes', background 'front-end developer'. Output: 12yo version 'a book's table of contents', dev version 'a B-tree map from column value to row pointer', then 'would an index on (last_name, first_name) serve a query filtering only on first_name? Why not?'
Prompt
Explain [CONCEPT] twice: once for a curious 12-year-old, once for someone with [YOUR BACKGROUND]. Then give me 3 questions that would reveal whether I actually understood your explanation — questions whose answers can't be faked.
Ruthless week triage
Planning
When to use it
The max-3 constraint forces real prioritization instead of the everything-is-important trap. Run it Monday morning on the raw brain-dump, before the day gets claimed.
Example
Input: 14 items. Output: DO TODAY: 'fix checkout bug, reply to client, send invoice'. DELETE: 'research new frameworks (no decision depends on it), reorganize bookmarks (cosmetic)' with one-line justifications.
Prompt
Here is my task list: [PASTE LIST]. Sort every item into: DO TODAY (maximum 3), SCHEDULE, DELEGATE-OR-AUTOMATE, DELETE. For anything in the delete pile, give one sentence on why dropping it is safe. Be ruthless — a list where nothing gets deleted is a failure of triage.