AI & Developer Tools

Prompt Engineering Cheat Sheet

Reusable, product-neutral techniques for giving context, constraints, examples and evaluation criteria to language models.

Jurisdiction
Global — general AI prompting
Applies to
OpenAI and Anthropic guidance checked 16 November 2026
Last verified
16 November 2026

At a glance

  • State the task, relevant context, desired output and constraints explicitly.
  • Examples are most useful when they reflect real inputs, edge cases and the desired format.
  • Evaluate prompts against representative cases instead of treating one good response as proof.

Reusable prompt frame

Task:
Context:
Inputs:
Constraints:
Required output format:
Quality criteria:
If information is missing or uncertain:
Examples (optional):

Technique map

NeedTechnique
Consistent structureSpecify headings, fields, schema or a short output example.
Domain contextSupply the policy, definitions, audience and decision being supported.
Fewer assumptionsTell the model to identify missing inputs and label uncertainty.
Complex source materialSeparate instructions, context and documents with clear labels or tags.
Reliable styleProvide a small set of relevant, diverse examples.
Quality controlDefine acceptance checks and test against normal and edge cases.

Safety and iteration

  • Do not paste secrets or personal data unless the product, account and task are approved for it.
  • Ask for citations when the task depends on changing facts, then verify the cited source.
  • Change one prompt element at a time and keep a small evaluation set.
  • Treat model output as a draft for high-impact legal, medical, financial or security decisions.

Practical answers

Frequently asked questions

Use these answers with the dated scope and authoritative sources shown on this page.

What is Prompt Engineering?

Reusable, product-neutral techniques for giving context, constraints, examples and evaluation criteria to language models.

What are the most important points about Prompt Engineering?

State the task, relevant context, desired output and constraints explicitly. Examples are most useful when they reflect real inputs, edge cases and the desired format. Evaluate prompts against representative cases instead of treating one good response as proof.

How should I use this prompt engineering cheat sheet?

Read the sections on Reusable prompt frame, Technique map, Safety and iteration, then use the Copy prompt template block where it fits your situation. Keep the inputs and assumptions with the copied result so somebody else can review it.

Which jurisdiction and effective date does this page cover?

It covers Global — general AI prompting and is labelled for OpenAI and Anthropic guidance checked 16 November 2026. Rules from another jurisdiction, tax year, product version or contract may produce a different answer.

What should I verify before acting on it?

Confirm exceptions and later changes in the linked material from OpenAI, Anthropic. The page was last verified 16 November 2026; provider terms, forms, thresholds, software features and local rules can change after that date.

Authoritative sources

Use the linked source for current instructions and exceptions before acting.

Educational reference only. It is not legal, tax, accounting, lending, insurance or investment advice.