Skip to content
>_MyGearHut

AI Prompt Generator Playbook

Operator playbook for building copy-paste prompts that work in ChatGPT, Claude, and Gemini — templates, quality gates, and a meta-prompt to generate three variants fast.

This page is the operator playbook — use the templates below today. Interactive UI lands next. Check your AI search readiness meanwhile.

What this page is

A prompt factory playbook for operators who ship with AI daily. The interactive generator UI ships later; today you get the same workflow we use internally — pick a job type, fill a brief, run the meta-prompt, paste the output into your model of choice.

No signup. No magic. Just structured prompts that reduce retries.


60-second workflow

  1. Pick a job category (marketing, coding, writing, business, sales, productivity, research, creative).
  2. Write a 3-line brief — audience, task, constraints.
  3. Paste the meta-prompt below into ChatGPT, Claude, or Gemini.
  4. Copy the best of three variants and run it as a fresh conversation (or system prompt for coding agents).
  5. Score the output with the quality gate checklist before you ship.

The brief template (fill this first)

CATEGORY: [marketing | coding | writing | business | sales | productivity | research | creative]
MODEL: [ChatGPT | Claude | Gemini]
AUDIENCE: Who reads or uses the output?
TASK: What must be true when you're done?
CONSTRAINTS: Length, tone, format, things to avoid, tools/stack, deadline.
CONTEXT: Paste bullets, data, or code snippets the model needs.
SUCCESS: How will you know the output is good enough?

Example:

CATEGORY: marketing
MODEL: Claude
AUDIENCE: B2B SaaS founders on LinkedIn
TASK: Write a launch post for our free GEO audit tool
CONSTRAINTS: Under 220 words, no hype words, one clear CTA, no em dashes
CONTEXT: Tool scans llms.txt, schema, meta — free, no signup
SUCCESS: A founder could paste this without editing

Meta-prompt: generate 3 ready-to-use prompts

Copy this entire block. Replace the bracketed section with your brief.

You are a senior prompt engineer. I will give you a job brief. Produce exactly THREE distinct, copy-paste-ready prompts I can send to [MODEL] to complete the task.

Rules for every prompt you write:
- Start with role + objective in sentence one.
- Include numbered steps or a clear output schema.
- Specify tone, length, and format (markdown table, JSON, code block, etc.).
- Add 2–3 "do not" rules to prevent generic fluff.
- End with a self-check line the model must answer before finishing.

Variation strategy:
- Prompt A: Fast draft — minimum tokens, good enough to ship today.
- Prompt B: Structured — tables, outlines, or step-by-step with examples.
- Prompt C: Expert — chain-of-thought hidden, edge cases, revision pass built in.

After the three prompts, add a 5-line "operator note" on which variant to use when.

MY BRIEF:
[PASTE YOUR BRIEF HERE]

Category starter prompts

Use these when you already know the job. Swap bracketed fields.

Marketing — campaign angle finder

You are a B2B growth operator. Given [PRODUCT] for [AUDIENCE], produce 5 campaign angles.
For each angle: hook (≤12 words), pain point, proof point, channel fit (LinkedIn/X/email), and one sample headline.
Exclude buzzwords: revolutionary, game-changing, unlock, leverage.
Output as a markdown table. Rank by expected CTR for [CHANNEL].

Coding — feature implementation spec

You are a staff engineer on [STACK]. Implement [FEATURE] with these constraints: [LIST].
Deliver: (1) file-level plan, (2) TypeScript interfaces, (3) pseudocode for core logic, (4) test cases as Given/When/Then, (5) risks.
Do not write full files unless asked. Flag assumptions with ASSUMPTION: prefix.

Writing — edit pass for operators

You are a technical editor. Rewrite the draft below for [AUDIENCE].
Goals: cut 20% length, active voice, one idea per paragraph, keep all facts.
Output: (A) revised text, (B) bullet list of cuts made, (C) one sentence summary.
Draft:
[PASTE]

Research — source-backed brief

You are a research analyst. Question: [QUESTION]
Deliver a one-page brief: executive summary (3 bullets), key findings, contradictions in sources, confidence level (high/medium/low) per claim, and "what would change my mind."
If you lack live data, say so and propose search queries I should run manually.
Cite source types, not fabricated URLs.

Model-specific tweaks

ModelAdd this line to any prompt
ChatGPT"Use concise sections. If uncertain, ask one clarifying question before proceeding."
Claude"Prefer structured markdown. Put the final answer first, reasoning after."
Gemini"Ground claims in the context I provided. Label inferences vs. stated facts."

Quality gate (before you ship)

  • Output matches the format you asked for (not a essay when you wanted JSON).
  • Constraints honored — length, tone, banned phrases.
  • No hallucinated stats, URLs, or API names — verify anything factual.
  • One clear next action for the reader or for you.
  • Good enough to paste without a second prompt — if not, iterate with: "Revise only section 2; keep everything else."

Status

Interactive UI (category picker, brief form, three variants, one-click copy) ships soon. Until then, the meta-prompt above is the generator — free, unlimited, commercial use OK.