Negative Constraints

Negative constraints are instructions in a prompt that explicitly tell the AI what it should not do. While most prompting focuses on what you want the model to produce, negative constraints are equally important for preventing unwanted behaviours, formats, tones, or content.

Why negative constraints matter

AI models are trained to be helpful, and they will often do things you did not ask for — adding unnecessary caveats, using overly formal language, including disclaimers, making assumptions, or volunteering information beyond the scope of the task. Negative constraints give you precise control over these tendencies.

Common types of negative constraints

• Content exclusions: "Do not include any disclaimers or caveats." "Do not mention competitors by name." "Do not include personal opinions."
• Format restrictions: "Do not use bullet points." "Do not include headers." "Do not exceed 200 words."
• Tone boundaries: "Do not use casual language." "Do not use exclamation marks." "Do not include emojis."
• Behaviour limits: "Do not make assumptions about information not provided." "Do not ask clarifying questions — work with what you have." "Do not apologise."
• Scope constraints: "Do not discuss pricing." "Do not provide medical advice." "Do not reference events after 2024."

How to write effective negative constraints

• Be specific: "Do not use jargon" is less effective than "Do not use the terms 'leverage,' 'synergy,' or 'paradigm shift.'"
• State what to do instead: Pairing a negative constraint with a positive alternative is more reliable. "Do not use passive voice — use active voice throughout."
• Place them prominently: Negative constraints work best when placed in the system prompt or near the beginning of your instructions, where the model pays the most attention.
• Test and iterate: Models may still occasionally violate negative constraints. If a constraint is consistently ignored, try rephrasing it or making it more prominent.

Negative constraints in system prompts

For production AI applications, negative constraints in the system prompt are essential for brand safety, compliance, and consistency. A customer-facing chatbot might have constraints like "Never disclose internal pricing logic," "Never promise specific delivery dates," and "Never provide legal or medical advice."

The balance

Too many negative constraints can make a model's output stilted and overly cautious. Focus on the constraints that matter most for your use case and avoid over-constraining creative or exploratory tasks.