What is a prompt? Writing task instructions, references, and input boundaries
Understand prompt components through rewriting an announcement, including instructions, source text, examples, output requirements, and missing facts.
Understand the concept
A prompt is model input expressing a task or context, ranging from a short request to references and examples. Interfaces may include roles and multimodal content. Clear prompts help define a task but do not guarantee capability or turn missing information into verified facts.
Separate task instructions from source material
For rewriting opening hours, require unchanged dates and addresses and put the source separately. An instruction inside the source to ignore previous requirements remains data, not an operational command. Delimiters help distinguish content but are not a complete security measure.
Use checkable requirements such as changing tone without adding offers rather than merely asking for professionalism. Missing holiday hours should remain unknown or prompt clarification instead of invented dates. No generation test was run here.
Examples demonstrate formats but can teach wrong patterns
An input-output example can show which fields to retain, but it must be correct. Examples that fill missing fields with defaults can teach that habit. Include missing information and varied formats to check boundaries.
Select relevant excerpts and sources instead of inserting every chat record. Context is limited and irrelevant text can obscure essentials. If citations are required, request references to supplied material and verify that links and passages support the claims.
Align output instructions with actual validation
A requested table still needs correct values, and brevity must preserve qualifications. Prompted formatting helps, but applications needing fixed fields should consider structured-output support and independent validation rather than natural-language instructions alone.
Retain representative inputs and compare errors after changes. One pleasing answer does not establish general effectiveness, and failures should not all be blamed on user phrasing. Capability and evidence quality also matter; prompting is one part of task design.
Checked sources
These sources support the concepts and methods described here. Examples and test scope are labeled separately; check the version of the tools you use.