What Constraints Do in a Prompt
Constraints are the boundaries and requirements you set on the AI's output — word count, tone, audience level, topics to include or exclude, formatting rules, and scope limits. They function like the walls of a channel: they don't slow the water down, they direct where it goes. Without constraints, the model makes its own choices about length, formality, breadth, and structure — and those choices rarely match what you actually need. Constraints aren't about making the task harder for the AI; they're about making the output immediately usable without reformatting or editing.
The Six Types of Constraints Worth Knowing
Length constraints (max 150 words, exactly 3 paragraphs) prevent sprawl and ensure the output fits its destination. Tone constraints (conversational, direct, no jargon, avoid passive voice) keep the voice consistent. Scope constraints (focus only on X, do not cover Y) prevent tangents. Audience constraints (assume the reader knows nothing about coding) calibrate complexity. Format constraints (bullet list, numbered steps, JSON) ensure the output integrates into your workflow. Exclusion constraints (do not include disclaimers, do not mention competitors) prevent the model from adding content that hurts rather than helps. Use the constraint types that matter for how the output will actually be used.
Length Constraints: the Most Impactful Constraint
The single most consistently useful constraint is a length limit. AI defaults to verbose output because verbosity is statistically associated with thorough, helpful responses in its training data. But in most real use cases — an email, a tagline, a tweet, a code comment, a product description — brevity is actually more valuable than coverage. Specifying 'max 100 words' or 'one sentence per bullet point' doesn't degrade quality; it forces the model to prioritize the most important content. For any task where you know the output will live in a specific destination, set a length constraint before submitting.
Tone and Voice Constraints
Without a tone constraint, the model defaults to a neutral, slightly formal register — which sounds like a textbook for casual contexts and like a blog post for professional ones. Tone constraints let you specify exactly the register you need: 'conversational but not casual,' 'confident and direct without being aggressive,' 'technical but accessible to non-engineers,' 'warm but professional, like talking to a trusted advisor.' The more you can describe the tone you want rather than just labeling it ('friendly'), the better the calibration. If you have examples of tone you want to match, paste them in — 'write in the style of this example' is one of the strongest tone constraints available.
Balancing Constraints Without Over-Restricting
The risk with constraints isn't that you'll use them — it's that you'll use too many that pull in different directions. 'Write in exactly 50 words, using 3 paragraphs, with a bullet list, two headers, and a 10-word summary at the end' creates contradictions the model can't fully satisfy. Focus on the two or three constraints that most directly affect whether the output is usable for your specific task. If the output still isn't right after applying the most important constraints, add one more at a time. Constraints should be additive, not contradictory.
Negative Constraints: What Not to Do
Telling the AI what to avoid is often more effective than telling it what to include. 'Do not use the word leverage,' 'do not include disclaimers or caveats,' 'do not use passive voice,' 'do not mention pricing' — these exclusions prevent specific, predictable failure modes. Negative constraints are especially useful when you've already gotten a draft that's mostly right but keeps including something you don't want. Rather than restructuring the entire prompt, add one negative constraint targeting the exact issue. This is a fast, targeted fix that works remarkably well.