This guide covers general practice for prompting models: separate instructions from context, delimit raw material, specify output structure, split complex tasks into steps and give a few examples, all organized around API calls. Its value is method and patterns, not an industry phrasebook, and it does not cover image or voice prompting. The docs follow model versions, so check the current page; the writing only works when combined with your own acceptance criteria.
What it does
- Instructions, context and output format in separate blocks
- Delimiters around raw material to process
- Complex tasks split into steps with intermediate results
- A few examples to constrain style and structure
- Examples built around API calls, ready for code
Who it is for
- Already using the API and needing stable prompt structure
- Results swing on similar tasks and a controllable method is needed
- Setting one prompt writing standard for a team
- Deciding whether a problem lies in the prompt or the model
How to get started
- Read the methods and pick the one or two you lack most
- Rewrite a prompt on real input, applying each technique
- Save the working version as a template with model and expiry notes
