Prompt engineering is one of the few AI skills with a genuinely free, genuinely official place to learn it properly. You do not need a paid course for the basics. Here is where to actually go, in an order that makes sense.
Start with Anthropic’s own prompt engineering docs
The official starting point is Anthropic’s prompt engineering overview. It is a living reference, not a one-off tutorial, and it covers:
- Writing clear, direct prompts
- Using examples effectively
- Structuring prompts with XML tags
- Role prompting (telling the model who to act as)
- Giving the model space to think before it answers
- Chaining prompts together for multi-step tasks
The overview page also links to the Claude Console’s own prompt generator, prompt templates and variables, and a prompt improver tool — useful once you have the concepts down and want to iterate faster.
Then do the hands-on version
The same official overview links out to a free, example-filled GitHub prompting tutorial. It is structured as nine short chapters covering the same ground as the written docs — basic prompt structure, being clear and direct, assigning roles, and the rest — but with exercises you actually run rather than just read. If you learn better by doing than by reading reference docs, start there and treat the written overview as something to come back to.
A practical order to work through it in
Read the overview page first for the concepts, in roughly this order: clarity and directness, then examples, then XML structuring for anything with multiple parts, then role prompting, then letting the model think before it answers, and finally prompt chaining once a single prompt stops being enough for the task. Each concept in the docs builds on the one before it, so skipping straight to chaining before the earlier ones are solid tends to produce prompts that are complicated without being effective.
Then go hands-on with the GitHub tutorial’s nine chapters, working through the exercises rather than just reading them. The exercises exist specifically because reading about a clear prompt and writing one that is actually clear are different skills, and the gap between them is where most of the real learning happens.
What this does not cover
This is prompting a model directly through its API or console. It does not cover the separate skill of building an automated system that chains prompts together with tools, retries and error handling — that is closer to what our AI Agents & Practical AI guides deal with, and it is a genuinely different skill from writing a single good prompt.
A note on other providers
Prompt engineering concepts transfer reasonably well between model providers — clarity, examples and structure help regardless of which model you are prompting — but the exact syntax (XML tags, system prompt conventions) is Anthropic-specific in places. If you are working mainly with a different provider’s model, check that provider’s own equivalent documentation rather than assuming every technique here maps across unchanged.
The honest limitation here
This guide points you to the right official material and explains what each part covers; it is not a substitute for working through the docs and the GitHub tutorial yourself, and we have not independently re-tested every technique described in Anthropic’s own documentation.