AI · Framework
Google's 13 Principles for Prompting LLMs
Thirteen practical rules for writing effective prompts, distilled from Google's guide
This cheat sheet distills key advice from Google's comprehensive prompt engineering guide into thirteen clear rules. It covers foundational techniques like few-shot prompting and chain-of-thought, parameter tuning, context framing, and multi-turn iteration. It serves as a quick reference for improving AI outputs without reading a full technical book.
The framework
- Understand the tool. A prompt is how you talk to a language model. The better the input, the better the output.
- You don't need to be a coder. Anyone can write prompts. But great prompting? That's a skill.
- Choose your model wisely. Gemini, GPT-4, Claude, they all have strengths. Know what you're working with.
- Zero-shot prompt = ask without examples. Few-shot = show examples first. Chain-of-thought = force it to "think" step by step. Use the right weapon for the job.
- You can control the vibe. Tweak the temperature (creativity), top-p, max tokens, and more to steer the output.
- Be specific. Tell it exactly what you want. Vagueness kills good output.
- Context is king. Add details, background, and constraints. Think of it like briefing a world-class assistant.
- Iterate like crazy. Good prompts aren't written, they're rewritten.
- Give examples. Want a specific tone, format, or structure? Show it.
- Build multi-turn prompts. Ask, follow-up, dig deeper. Treat the model like a convo, not a vending machine.
- Avoid traps: Too vague = bad output; Too long = confusion; Too complex = model gets lost; Biased wording = biased results.
- Prompts aren't one-size-fits-all. Customize by task: writing, summarizing, coding, customer support, etc.
- Prompt engineering isn't optional. It's how you unlock superpowers with AI.
What it's for
- Structure prompts for complex reasoning tasks using few-shot or chain-of-thought methods
- Troubleshoot low-quality AI outputs by identifying vagueness, length, or bias issues
- Guide parameter selection such as temperature and token limits for specific workflows
- Design multi-turn conversational interactions for iterative problem solving
Key points
- Overview of core prompting styles including zero-shot, few-shot, and chain-of-thought
- Guidance on selecting models and configuring generation parameters like temperature and top-p
- Principles for supplying context, constraints, and explicit output formatting
- Techniques for iterative refinement and conversational multi-turn prompting
- Checklist of common prompting mistakes such as over-complexity and biased phrasing
Pricing
Free
Details
- Type
- Framework
- Category
- AI
- Topic
- Prompt Engineering
- Open source
- No
- Platforms
- Web
- For
- Anyone using LLMs for writing, coding, research, or automation
- Alternative to
- OpenAI Prompt Engineering Guide, Anthropic Prompt Engineering Interactive Tutorial
- Tags
- prompting, llm, ai, framework, cheat sheet, generative ai
Why it was saved
- @gregisenberg: A complete cheat sheet for writing better prompts: model choice, prompting techniques, settings, specificity, context, iteration, and common traps.
Related
A comprehensive guide to prompt engineering techniques and LLM parameters by Google.
Prompt formula and best practices for realistic AI video generation
Curated collection of 50+ tested prompts for Nano Banana Pro image generation
Prompt framework to build structured conversational scripts for voice agents.
Turn rough task descriptions into structured, production-ready AI prompts automatically
Curated repository of prompts and styles for the Gemini Nano Banana model
Updated