Task Patterns for Prompt Engineerings
Task Patterns are tested prompt shapes for common jobs: summarising, classifying, extracting fields, rewriting, translating and answering questions. Structured output is the key skill here, asking the model to return JSON that follows a schema so code can parse it, ideally with the provider's schema-constrained mode and validation.
Stage: Applied Prompting · Part of the Prompt Engineering roadmap · Last reviewed 19 November 2026
What to learn in Task Patterns
- Write an extraction prompt that returns named fields as JSON
- Constrain a classifier to a fixed list of allowed labels
- Use a JSON schema or structured output mode when the API supports it
- Validate model output in code and retry on a parse failure
- Ask for a fallback value when information is missing instead of guessing
Task Patterns interview questions
How do you get reliable JSON from a language model?
Provide the schema and an example, use the API's structured output or JSON mode where available, keep temperature low, and validate the result in code. On failure, retry with the error message. Never assume the output is valid.
How do you reduce made-up values in an extraction task?
Tell the model to use null or unknown when a field is absent, require quotes from the source text as evidence, and check the extracted values against the document in code. Explicit permission to say not found cuts guessing.
Where Task Patterns fits in the Prompt Engineering roadmap
Task Patterns belongs to the Applied Prompting stage (beginner level), which takes about 12 hours. Real tasks.
Other units in this stage
Back to the full Prompt Engineering roadmap