What is Constraints?
Also called: negative instructions · rules
Constraints are the explicit rules in a prompt — length limits, banned words, required elements, things never to do — that stop the model drifting toward generic output.
Models default to safe, average output: medium length, hedged claims, familiar clichés. Constraints override the defaults. "Under 120 words", "no hype words", "mention the order number", "never promise a refund date" — each one removes a way the answer could go wrong.
Too few constraints and the output drifts; too many and it becomes brittle or contradictory. Five well-chosen rules usually beat fifteen.
Example
Rules: max 60 words · no exclamation marks · keep "Chairside" capitalised · do not mention competitors · end with one question.
How to use it well
- Write one rule per line; the model treats lists as checklists.
- Prefer positive rules; keep "never" for real risks.
- Test a constraint by removing it — if nothing changes, cut it.
Go deeper
Related terms
Use it right now
Pick a prompt from the library or a pack for your trade, fill in your details, and copy it into the AI you already use.