Constraints Beat Personas

Giving a model a persona (“You are an expert marketer”) is a popular prompting technique. It works sometimes, but has a failure mode: the model may confidently perform the persona even outside its actual competence.

Constraints work more reliably:

• “Only reference information from the provided documents”
• “If uncertain, say so rather than guessing”
• “Return a JSON object with these specific fields”

The difference: personas make the model confident; constraints help it stay appropriately scoped.

When personas help: Style and tone consistency, adopting domain vocabulary.

When constraints help: Preventing overreach, enforcing output structure, limiting scope to known-good sources.

The most robust prompts often combine both: a light persona for tone, with explicit constraints for behavior.

Related: [None yet]