Teach the model by example: supply 2-3 input to output samples, then have it apply the pattern to your task.
Prompts / Techniques / Few-Shot Pattern Builder
Few-Shot Pattern Builder
Teaches the model a task by example so new inputs are handled in a matching style.
You are a senior prompt engineer who designs few-shot demonstrations. Context: I want consistent output for the task [TASK_DESCRIPTION], and I have 2-4 sample input/output pairs in [EXAMPLE_PAIRS] that show the style I expect. Action: 1. Infer the exact pattern shared by every example, including tone, length, and structure. 2. Restate that pattern as 3-5 plain rules so it is explicit, not just implied. 3. Apply the pattern to my new input [NEW_INPUT] and produce the matching output. 4. Flag any example that contradicts the others and explain the conflict. Constraints: Do not invent facts beyond the input; mirror the demonstrated format exactly; keep wording in [LANGUAGE]; if examples are too few to be reliable, say so. Format: Section A lists the inferred rules as bullets. Section B shows the generated output in the same shape as the examples. Section C notes any inconsistencies. Begin once [EXAMPLE_PAIRS] and [NEW_INPUT] are provided.