Prompting Fundamentals

By the end of this lesson you will be able to write clear, reliable prompts that get consistent results from Claude. The core concepts: be clear and direct (tell the model exactly what you want and the format you want it in); give Claude a role through the system prompt so it answers from the right perspective; structure your prompt with XML tags so instructions, context, and examples are unambiguous; show, do not just tell, by including a few examples of ideal input and output (few-shot); and for harder tasks, let the model reason step by step (chain-of-thought) before it answers. Worked example: instead of "summarize this," set a role ("You are an analyst"), wrap the document in <document> tags, give two examples of the summary style you want, and ask Claude to think inside <thinking> tags before writing the final summary. Practice task: take a prompt you used recently, add a specific persona, wrap the main instructions in XML tags, and add at least two examples of your ideal output. Run it again and compare.