Controlling Tone and Output Format

Tone is the voice the model uses; format is the shape of the answer. You tell the model both.

How to control tone (simple):

• Say the voice and who the reader is. Example: “Write in a warm, conversational tone for busy parents.”
• Give one example sentence if you want a very specific voice to copy.

How to control format (simple):

• Be exact about structure. Example: “Return a JSON object with keys: title (string), summary (string, max 140 chars), tags (array of strings).”
• Tell the model to include only the machine-readable block: “Provide only the JSON object and nothing else.” Use code fences if you want a clean code block.

Example prompt asking for JSON (copyable):

You are a data extractor. Read the text and return a JSON object with keys:
  - "title" (string)
  - "summary" (string, max 140 characters)
  - "tags" (array of strings)

Return only the JSON object and nothing else.

Why this works: the model knows who to be, what to output, and the exact format. That reduces errors when your app parses the result.

Quick tips:

• Use short examples to show exactly how the output should look.
• Prefer exact counts and types (e.g., “3 bullets”, “max 120 chars”).
• When building programs, prefer strict machine-readable formats (JSON, CSV) and validate outputs on the server.