The Ultimate Prompt Engineering Cheat Sheet (2026)

title: "The Ultimate Prompt Engineering Cheat Sheet (2026)" slug: "49-prompt-engineering-cheat-sheet" description: "One-page prompt engineering cheat sheet covering frameworks, parameters, models, common patterns. Bookmark this. CRAFT, CoT, few-shot, JSON, RAG." publishedAt: "2026-07-19" updatedAt: "2026-07-19" postNum: 49 pillar: 5 targetKeyword: "prompt engineering cheat sheet" keywords:

• "prompt engineering cheat sheet"
• "prompt cheat sheet"
• "ai prompt reference"
• "llm cheat sheet" ogImage: "https://prompt-architects.com/og/49-prompt-engineering-cheat-sheet.png" author: name: "Nafiul Hasan" role: "Founder, Prompt Architects" url: "https://prompt-architects.com/about" ctaFeature: "generator" related: [41, 1, 6] faq:
• q: "Is this a downloadable PDF?" a: "Yes — print this page or save as PDF (Cmd+P → Save as PDF). The structure is designed for one-page printing on letter or A4. Hyperlinks remain clickable in the PDF for digital reference."
• q: "What's the most useful single thing on this cheat sheet?" a: "The CRAFT framework (Context, Role, Action, Format, Tone). Lifts quality 62% over unstructured prompts on average. Covers 80% of daily AI tasks. Master it first."
• q: "Does this cover Claude and Gemini too?" a: "Yes — the frameworks (CRAFT, CoT, CARE) and patterns (JSON, structured output, few-shot) apply across ChatGPT, Claude, and Gemini. Some model-specific parameters differ; noted where relevant."
• q: "How often does this cheat sheet need updating?" a: "Quarterly. Models update; new patterns emerge. The core frameworks are stable; specific model-tier picks shift. Re-check when a major model version ships."
• q: "Is this enough to learn prompt engineering?" a: "It's a reference, not a tutorial. Pair with practice — apply each framework to 5-10 real tasks. After 50 deliberately-structured prompts, you'll internalize the patterns and won't need the cheat sheet for daily work."

TL;DR: One-page reference covering frameworks, parameters, model picks, common patterns. Print, bookmark, or save as PDF. Updated quarterly.

The 5 frameworks (covers 80% of tasks)

CRAFT template (memorize this)

[CONTEXT] [Background — 1-2 sentences about your situation]
[ROLE] Act as a [specific role with N years experience]
[ACTION] [Single specific verb + noun]
[FORMAT] [Output shape: list, table, JSON, paragraph length]
[TONE] [3-5 voice attributes]

Sampling parameters quick reference

Model picks (April 2026)

Top 10 leverage moves

1. Add a role. "Act as a [specific role with N years experience]." +18% success rate.
2. Specify format. "Numbered list, ≤8 words each" or "JSON matching schema". +21% success rate.
3. Add 1-2 examples (few-shot). +27% success rate. Halves rework on repeated patterns.
4. Specify audience. "Writing for [audience with assumed knowledge]." +14% success rate.
5. Add 2-3 voice attributes in tone. +9% success rate. Cap at 5; more confuses model.
6. Use Chain-of-Thought for reasoning ("Let's think step by step"). +30-71% on math/logic.
7. Cap length explicitly. "≤200 words". Prevents bloat.
8. Single task per prompt. Multi-task prompts fail 2.4× more. Chain instead.
9. Use specific verbs. "Outline", "summarize", "extract" beat "help me with".
10. Set --raw on Midjourney photo prompts. Default house aesthetic kills realism.

JSON prompt template

For production AI / structured output:

{
  "role": "[who the AI should be]",
  "task": "[specific verb + noun]",
  "input": "[your data]",
  "constraints": {
    "[rule_1]": "[value]",
    "[rule_2]": "[value]"
  },
  "output_schema": {
    "[field_1]": "[type or example]",
    "[field_2]": "[type or example]"
  }
}

Respond as JSON matching output_schema. No prose, no code fences.

Common mistakes (avoid these)

1. Vague verbs ("help me with X")
2. Missing format (model picks prose by default)
3. Multi-task in one prompt (chain instead)
4. 5+ voice attributes (model averages)
5. Trusting first output blindly (always read for accuracy + AI tells)
6. Hard-coded variables (use {{placeholders}} for repeated patterns)
7. No re-test after model updates (modifier weights drift)

Decision tree: which framework?

What's your task?
├── General task → CRAFT
├── Math, code, logic → Chain-of-Thought
├── Have a sample of desired output → CARE
├── Quick simple Q&A → RTF or TAG
├── Custom classification or extraction → Few-shot
├── Repetitive pattern at scale → Few-shot + JSON mode
└── Multi-step with brand voice → CRAFT + CoT + few-shot

When to use which

Default order: prompting → RAG → fine-tuning. Skip ahead only when previous step demonstrably fails.

Production AI checklist

• System prompt sets role + voice + format + refusal rules
• User prompt has the specific task + data
• Output schema validated (Zod / Pydantic / JSON Schema)
• Tool calls require human confirmation for destructive actions
• Input filtering for top injection patterns
• Treat LLM output as untrusted (don't eval, don't render raw HTML)
• Cost + latency monitoring (Helicone, PromptLayer, LangSmith)
• A/B test prompts against versioned baselines
• Monthly red-team test for prompt injection

Image prompt cheat sheet (Midjourney v7)

Video prompt cheat sheet (Veo 3)

6-part structure:

1. Subject — specific descriptors (age, hair, clothing)
2. Action — what they're doing
3. Scene — where, when, weather
4. Camera — framing, lens, movement (2-3 modifiers)
5. Lighting — source, direction, mood
6. Audio — dialogue, ambience, score (Veo 3 syncs natively)

Resources for going deeper

• OpenAI prompt engineering guide: https://platform.openai.com/docs/guides/prompt-engineering
• Anthropic prompt library: https://docs.anthropic.com/en/prompt-library
• Google Prompt Engineering Guide: https://cloud.google.com/discover/what-is-prompt-engineering
• Prompting Guide (community): https://www.promptingguide.ai/
• Prompt Architects blog: dozens of deeper posts on each framework above

Quick wins (ship today)

1. Save 10 prompts as templates with {{variables}}.
2. Add CRAFT structure to next 5 prompts. Compare output to unstructured.
3. Try Chain-of-Thought on one math/logic problem. Note the lift.
4. Set up structured output mode for one production extraction task.

Print / save as PDF

This page is designed for one-page printing. Cmd+P → "Save as PDF" or your browser's print-to-file. Hyperlinks remain clickable in PDF.

For deeper learning, follow the Related Posts links throughout this site. Each framework has its own deep-dive post.

Subscribe for updates

This cheat sheet updates quarterly when major model versions ship. Subscribe via RSS at /feed.xml to get notified — or just bookmark this page and re-check when a new version of GPT, Claude, Gemini, or Midjourney ships.

Tools that ship these frameworks as one-click presets (Prompt Architects) accelerate daily work. The skill underneath is what matters; the cheat sheet is a memory aid for it.