Write the Constitution Before the Clever Prompt
Define the small set of system-prompt instructions that stabilize a custom assistant across many requests.
Reliability comes less from clever phrasing than from putting the right rules in the right layer. What belongs in the top layer Put stable behavior in the system prompt: role, audience, output format, source precedence, and how the assistant should respond when evidence is missing. These are the instructions that should survive every new user message. Why this works A model handles ambiguity by guessing. A good system prompt removes whole categories of guessing before the conversation starts. That lowers variance across turns, especially when different coworkers ask the assistant in different ways. What does not belong there Do not…
Sign up free — one personalized lesson every day, matched to your role and goals.
Already have an account? Sign in