Skip to content
Home › Prompt Engineering & Structured Output

Prompt Technique Taxonomy

Prompt Engineering & Structured Output

Definition

The progressive ladder of prompting techniques taught by Anthropic's interactive prompt-engineering tutorial (anthropics/courses), ordered from foundational structure to complex, composed prompts.

Key points

  • The ladder (tutorial chapter order):
    • Basic prompt structure — the shape of a good prompt.
    • Being clear and direct — say exactly what you want, unambiguously.
    • Assigning roles (role prompting) — give Claude a persona/role to steer tone and expertise.
    • Separating data from instructions — keep the task instructions distinct from the input data (commonly via XML tags).
    • Formatting output & speaking for Claude — control output format; prefill the assistant turn to steer/skip preamble (see Structured Outputs).
    • Precognition / think step by step — chain of thought (CoT) reasoning before the answer.
    • Using examples — few-shot prompting (see Few-Shot Prompting).
    • Avoiding hallucinations — techniques to reduce fabricated content (give an out, ground in provided data, ask for evidence).
    • Building complex prompts — compose the above for real industry use cases (chatbots, legal, financial, coding).
  • Appendix / advanced techniques: prompt chaining, tool use, and search & retrieval.
  • Applied practice (Real World Prompting): a systematic prompt-engineering process applied to medical, call-summarization, and customer-support-bot walkthroughs.
  • The taxonomy is a ladder, not a menu — start with clarity/structure, escalate to CoT / few-shot / complex composition only as the task demands.

Why it matters for the exam

  • The Prompt Engineering domain expects you to name and order these techniques and match each to the problem it solves.
  • Distinguishing role prompting vs. CoT vs. few-shot vs. data/instruction separation is a common question shape.

Common gotchas

  • Conflating role prompting (persona) with CoT (step-by-step reasoning) — different levers.
  • Jumping straight to complex/composed prompts before ensuring the prompt is clear and direct.
  • Treating "speaking for Claude" (prefill) as separate from output formatting — the tutorial groups them.

See also

Sources

Referenced by