Controls how much scaffolding wraps your prompt. Minimal = just the task. Moderate = role + format + constraints. Full = complete system prompt with all sections.

Assigns the model a professional persona. This colors word choice, reasoning style, and rigor.

Adjusts vocabulary, assumed expertise, and framing based on who the output is for.

Sets the emotional register and formality of the writing — word choice, sentence rhythm, and how assertive or warm the output feels. Most impactful for emails, briefs, and reports.

Controls creativity vs precision, and whether to enable step-by-step reasoning.

Controls whether the model asks clarifying questions first, or self-reviews its draft.

Explicitly tells the model the structure to use. Without this, the model guesses — and often guesses wrong. Selecting multiple formats is allowed (e.g. Executive summary + Bulleted list).

Sets expectations on how much the model should write. Most models over-explain by default — constraining length keeps outputs focused. The word count field overrides the label if you need precise control.

Adds explicit verification requirements. Best for research, analysis, and high-stakes outputs. Strawman asks the model to argue against its own conclusion — surfacing weaknesses, blind spots, and counterarguments you might have missed. Steelman does the opposite: it asks the model to strengthen its weakest points, filling gaps in logic and reinforcing the argument before it reaches a reader. Both catch problems before the output leaves your hands.

Background information the model needs to write accurately. Without it, the model fills gaps with assumptions or generic content. Paste meeting notes, prior emails, data, specs — the more relevant context, the more grounded the response.

Control whether the model uses web search, only provided context, or its own knowledge.

Request structured JSON output matching a specific schema. Useful for API integrations, data pipelines, or any time you need the model to return consistently shaped data rather than free-form text.

Showing the model what good output looks like is often more effective than instructions. Even one example communicates tone, length, and structure.

Showing what NOT to do is one of the highest-leverage prompt techniques. Often more effective than positive instructions alone.

Defines what the model must NOT do. Explicit exclusions are often more reliable than positive instructions — "Do not include pricing" is clearer than "keep it high-level." Good for enforcing house style, preventing recurring mistakes, or limiting scope.