AI Assistant Summary: Anthropic publishes a free, comprehensive Interactive Prompt Engineering Tutorial on GitHub at github.com/anthropics/prompt-eng-interactive-tutorial (35,600+ stars). It’s 9 chapters plus an appendix, organized into Beginner (chapters 1-3: prompt structure, clarity, role assignment), Intermediate (chapters 4-7: separating data from instructions, formatting, step-by-step thinking, using examples), and Advanced (chapters 8-9: avoiding hallucinations, building complex prompts for chatbots, legal/financial services, and coding) plus an appendix on prompt chaining, tool use, and retrieval. Available as Jupyter notebooks or interactive Google Sheets. Uses Claude Haiku as the demo model with examples for Sonnet and Opus. This guide walks through what each chapter teaches, who it’s for, and how to actually complete it.
Prompt engineering is the single highest-ROI skill for working with Claude (or any LLM). Anthropic publishes their own official course on the topic, completely free, on GitHub. With 35,600+ stars, it’s one of the most respected resources in the field. This guide explains what’s in it, who should take it, and how to actually finish it without giving up halfway.
What is the course?
- Location: github.com/anthropics/prompt-eng-interactive-tutorial
- Format: Jupyter notebooks (primary) + interactive Google Sheets (alternative, often easier for non-developers)
- Length: 9 chapters + appendix, structured in 3 difficulty tiers
- Cost: Free. You need an Anthropic API key to run examples (free starter credits cover the whole course).
- Time to complete: ~6-10 hours if you do every exercise; ~3 hours if you skim.
- Audience: Developers and non-developers who want to write better prompts.
- Model used: Claude Haiku as the demo (cheapest); references Sonnet and Opus for harder tasks.
The 9-chapter structure
| Tier | Chapter | What you learn |
|---|---|---|
| Beginner | 1. Basic Prompt Structure | The Messages API format; system vs user roles |
| Beginner | 2. Being Clear and Direct | Why specificity beats “magic phrases” |
| Beginner | 3. Assigning Roles | The system prompt as persona shaper |
| Intermediate | 4. Separating Data from Instructions | XML tags as clean boundaries |
| Intermediate | 5. Formatting Output | Speaking-for-Claude technique; structured outputs |
| Intermediate | 6. Precognition (Step-by-Step Thinking) | Chain of thought without manual extended thinking |
| Intermediate | 7. Using Examples (Few-Shot) | When and how to include examples |
| Advanced | 8. Avoiding Hallucinations | Techniques to keep Claude grounded |
| Advanced | 9. Building Complex Prompts | Industry walkthroughs: chatbot, legal, financial, coding |
| Bonus | Appendix | Prompt chaining, tool use, search & retrieval |
1-on-1 Coaching
Claude AI Crash Course
1-hour private video session with James. Work through the parts of Anthropic’s prompt course that matter for your specific work, build prompt templates that fit your workflow, and leave with a saved library you can keep using. Faster than self-paced for most people.
Group Format
AI Workshops for Teams
Team-format workshops walking through Anthropic’s prompt course as a group, building team-specific prompt templates, and standardizing workflows. Best for 3+ people.
Beginner tier: Chapters 1-3
Chapter 1: Basic Prompt Structure
Introduces the Anthropic Messages API: how a prompt is actually structured (system prompt, user message, optional assistant prefill). Most beginners don’t realize Claude works in turn-based messages rather than a single text blob — this chapter fixes that mental model.
Why it matters: If you’re going to use the API at all, this is the floor. The structure determines everything that comes next.
Chapter 2: Being Clear and Direct
The single most important prompt-engineering principle: clarity beats cleverness. The chapter shows side-by-side examples of vague vs. specific prompts and the dramatic output differences. Teaches the “Goldilocks principle” — not too much, not too little, just the right amount of context.
Why it matters: Most prompt failures aren’t because the model isn’t smart enough — they’re because the prompt didn’t tell the model what was wanted. This chapter teaches you to fix that.
Chapter 3: Assigning Roles
The role/persona pattern: telling Claude “you are a [role]” in the system prompt shifts every response in that role’s direction. Examples: “you are a copyeditor,” “you are a Socratic tutor,” “you are a senior software engineer reviewing code.”
Why it matters: Role assignment is the cheapest, highest-leverage prompt change. Same user message + different role = completely different output. This unlocks 80% of system-prompt usefulness in one technique.
Intermediate tier: Chapters 4-7
Chapter 4: Separating Data from Instructions
The XML-tag pattern: wrap your data in clear tags (<document>, <email>, <feedback>) so Claude knows what’s input and what’s instruction. Without this, prompts often confuse Claude on long inputs.
Why it matters: Claude was specifically trained to understand XML tags as boundaries. This is a Claude-specific technique that doesn’t always transfer to other models, but it works extremely well here.
Chapter 5: Formatting Output (Speaking for Claude)
How to make Claude produce specific output formats reliably: JSON, structured lists, exact field schemas. The “speaking for Claude” technique — prefilling the assistant response with “{” — forces JSON output without any other prompting.
Why it matters: Production applications need structured outputs. This chapter is how you get them reliably without resorting to fragile parsing.
Chapter 6: Precognition (Step-by-Step Thinking)
Chain-of-thought prompting: asking Claude to “think step by step” before answering. The chapter shows the dramatic accuracy improvement on reasoning tasks. Also touches on when to use this vs. when to use formal extended thinking (available on Sonnet 4.6 and Haiku 4.5).
Why it matters: One sentence in your system prompt can change a 60% accuracy task into a 90% accuracy task. The cost is a few extra output tokens.
Chapter 7: Using Examples (Few-Shot Prompting)
When and how to include examples in your prompt: zero-shot, one-shot, few-shot. The chapter shows that examples beat almost any other technique for narrow tasks — classification, formatting, style matching. Also addresses common mistakes: too many examples, contradictory examples, examples that anchor on irrelevant patterns.
Why it matters: “Show, don’t tell” works for AI. If you can give 2-3 examples of what good output looks like, you almost don’t need other prompt instructions.
Advanced tier: Chapters 8-9
Chapter 8: Avoiding Hallucinations
Techniques for keeping Claude honest about what it knows and doesn’t know: explicit “say I don’t know if uncertain,” citation requirements, source-tied output, and the powerful “use only the provided document” instruction that locks Claude to a specific input.
Why it matters: Every production AI application eventually has to deal with hallucinations. This chapter is the playbook for reducing them substantially without giving up the model’s broader capabilities.
Chapter 9: Building Complex Prompts (Industry Use Cases)
The capstone chapter. Walks through full prompt design for four real industries:
- Customer service chatbot — persona, fallback handling, escalation triggers
- Legal services — clause analysis with citation requirements and disclaimers
- Financial services — analysis with regulatory hedging and uncertainty flagging
- Coding assistant — review patterns, suggested-vs-final structure, when to ask for clarification
Why it matters: Real production prompts combine every technique from chapters 1-8. This chapter shows what that looks like in practice. Copy any of the four templates and customize for your domain.
Appendix: Beyond Standard Prompting
The appendix introduces three patterns beyond single-prompt work:
- Prompt chaining — breaking complex tasks into a sequence of smaller, more reliable prompts.
- Tool use — giving Claude functions to call. Covered in more depth in the official tool use documentation and the Cookbook.
- Search and retrieval (RAG) — how to combine prompts with external knowledge bases.
The appendix is a teaser more than a full curriculum. For each topic, the official docs and Cookbook examples go deeper.
How to actually finish the course
The drop-off rate on self-paced courses is brutal — most people start, fail to budget time, and abandon by chapter 4. Patterns that work:
- Block 90-120 minute sessions. Two chapters per session is realistic. Don’t try to do it 15 minutes a day — the context-switching cost is too high.
- Use the Google Sheets version if you’re not a developer. The Sheets format is more user-friendly than the Jupyter notebooks. You don’t need a coding environment.
- Use the Jupyter version if you are a developer. Run it in Google Colab or your local Python environment. Each chapter is a single notebook.
- Apply each chapter to a real prompt you actually need. If you have a work prompt that’s been frustrating, use the chapter’s techniques to fix it. Real-world application sticks better than abstract examples.
- Use the answer key. Anthropic provides a separate Google Sheet with answers to the exercises. Check yourself after each exercise.
- Skip Chapter 9 on first pass if you’re impatient. The industry walkthroughs are excellent but long. Chapters 1-8 + Appendix gives you 80% of the value in 60% of the time.
When to take this vs. other prompt resources
| Resource | Best for |
|---|---|
| Anthropic’s Interactive Tutorial | Anyone serious about prompting Claude specifically |
| Anthropic’s prompt engineering docs | Quick reference; specific pattern lookup |
| Anthropic Academy | Broader AI fluency (Claude + general concepts) |
| Learn Prompting (learnprompting.org) | Multi-model prompt engineering |
| DAIR.AI Prompt Engineering Guide | Academic-research-flavored prompting |
| Our Best Claude Prompts post | 50 ready-to-copy templates |
If you only do one: the official Anthropic tutorial is the highest-yield investment. It’s authored by the people who built Claude, runs on real Anthropic models, and covers everything except the agentic patterns (which the Cookbook handles).
Frequently asked questions
Do I need coding skills to take this course?
Not strictly. The Google Sheets version is fully usable without code. You’ll see formulas calling the Claude API, but you don’t need to modify them — you only modify the prompts in designated cells. The Jupyter version is friendlier for developers.
Will the techniques transfer to ChatGPT or Gemini?
Most do; some don’t. The general principles (clarity, examples, role assignment, chain-of-thought) transfer cleanly. Claude-specific patterns (XML tags, “speaking for Claude” prefill) work in modified forms on other models but aren’t optimal. For multi-model production work, learn this course first, then read each vendor’s specific guide.
Is the course still current?
Anthropic updates the repository periodically. The model used as the demo is older (Haiku 3 in some chapters), but the principles are model-agnostic. Newer models (Sonnet 4.6, Haiku 4.5, Opus 4.7) follow the same prompting patterns. Watch the repo for major updates.
How does this compare to Anthropic Academy’s courses?
Anthropic Academy has broader AI fluency content (good for educators, students, general users). The Interactive Tutorial is laser-focused on prompt engineering for builders. They complement each other — Academy first if you’re new to AI, Interactive Tutorial if you specifically need to write better prompts.
Can I share my completed work?
Yes — fork the GitHub repo, push your notebook with your answers, and share the link. Many job applicants in AI-related roles include completed Anthropic tutorial work as part of their portfolio.
How long do the API credits last for the whole course?
The whole course on Haiku 4.5 typically costs $0.50-2.00 in API tokens. The free starter credits on signup cover the entire tutorial with room to spare. Our free tier guide goes deeper on the credit allocation.
Is there a video version?
Anthropic’s official YouTube channel has talks on prompt engineering (search “Anthropic prompt engineering”), but the course itself is text/notebook format. Third-party YouTube walkthroughs exist with varying quality — treat them as supplemental.
Keep getting better at prompting
New prompting patterns & Claude updates, daily
Free daily newsletter — one story, one tool, one tip. Plain English, no jargon. One issue per day.
Free forever. Unsubscribe anytime.
You may also like
- Anthropic Academy: Free Courses to Learn Claude
- Anthropic Cookbook: 10 Code Examples From the Official Repo
- Anthropic Documentation: 12 Pages to Bookmark
- Anthropic Workbench: Step-by-Step First Use
- Claude Prompt Engineering: Advanced Techniques That Work
- Best Claude Prompts: 50 Examples That Actually Work
- Claude Opus vs Sonnet vs Haiku: Which to Use in 2026
- Anthropic’s Free Tier
Sources
- Anthropic. “Interactive Prompt Engineering Tutorial.” github.com/anthropics/prompt-eng-interactive-tutorial (verified May 14, 2026)
- Anthropic. “Prompt engineering documentation.” platform.claude.com/docs/build-with-claude/prompt-engineering
- Anthropic. “Tool use.” platform.claude.com/docs/build-with-claude/tool-use
- Anthropic. “Cookbook.” github.com/anthropics/anthropic-cookbook
Two ways to go further
The AI Prompt Library
1,000+ ready-to-use prompts for Claude, ChatGPT, and Gemini. Stop staring at a blank box.
Get it for $39 →2-Hour Live AI Crash Course
A private, beginner-friendly session across Claude, ChatGPT, Gemini, and the wider landscape.
Book for $125 →