Use Case & Objective
Tips
  • Text Generation: audience, voice, sections, length, calls-to-action.
  • Summarization: bullet count/length, must‑include facts, no new info.
  • Classification: label definitions, tie‑break rule, JSON output with rationale.
  • Information Extraction: required fields, null for missing, strict JSON schema.
  • Closed‑book Q&A: scope limits, uncertainty/refusal style.
  • RAG Q&A: answer ONLY from context, require citations (chunk IDs).
  • Reasoning/Planning: limit steps, final one‑line answer.
  • Coding: language/version, style guide, tests, return minimal diff.
  • Data/SQL: include DB schema, SELECT only, LIMIT rows.
Context & Constraints
What to include
  • Context: paste only relevant facts, policies, or retrieved chunks. Use clear delimiters.
  • Constraints: banned topics/phrases, must‑include items, tone rules, refusal policy.
  • RAG: include chunk IDs and toggle “ONLY use context” and “Require citations”.
Output Format
JSON schema example
  • Use strict schemas; set missing fields to null.
{
  "type": "object",
  "properties": {
    "title": {"type":"string"},
    "bullets": {"type":"array","items":{"type":"string"}},
    "sentiment": {"type":"string","enum":["positive","neutral","negative"]}
  },
  "required": ["title","bullets","sentiment"],
  "additionalProperties": false
}
Few‑Shot Examples (optional)
How to use examples
  • Add 1–3 concise input→output pairs that match your desired format and tone.
  • Keep examples short and focused on structure.
Advanced