Constraints are limits you set in the prompt that define the boundaries of acceptable output. Without them, the model defaults to a verbose, neutral register at an assumed reading level for an assumed audience. That default is rarely exactly what you need.
Constraints narrow the model's output space from "anything plausible" to "exactly what this use case requires." Each constraint eliminates a class of outputs you would have had to filter or rework after the fact.
Length — "In under 100 words." / "In exactly three sentences." / "In five bullet points." — controls how much the model produces
Tone — "Use a professional, formal tone." / "Keep it conversational and friendly." / "Be concise and direct." — sets the register and style
Audience — "Assume the reader is a complete beginner." / "Write for a senior engineer." — calibrates reading level and assumed knowledge
Scope — "Focus only on Python 3." / "Do not discuss edge cases." / "Answer only from the provided text." — defines what topics are in and out
Vocabulary — "Use only simple English." / "Write in second-person." / "Avoid jargon." — sets specific language requirements
"Keep it short."
"Short" means different things to different models and different contexts. You will get outputs ranging from one sentence to three paragraphs.
"Be simple."
"Simple" is interpreted relative to what the model assumes is normal. It may still include vocabulary your audience does not know.
"In under 50 words."
A hard number. The model can check its own output against it. You get consistent length control.
"Use only words a 10-year-old would know. No technical vocabulary."
Concrete audience + explicit prohibition. The model has a clear target to optimize toward.
Use numbers, not adjectives — "under 100 words" not "short"; "exactly 5 bullet points" not "a few points"
Use negation to prevent defaults — "Do not include code examples" prevents the model from adding examples it would insert by default
Stack constraints freely — multiple constraints apply simultaneously and narrow the output space further; they do not conflict unless logically contradictory
Place constraints near the task instruction — constraints buried at the end of a long prompt are less reliably followed than those adjacent to the task
Contradictory constraints produce unpredictable output. "Explain this in technical detail" and "Use no technical vocabulary" are contradictory — the model must choose one to follow and may satisfy neither cleanly. Before sending a prompt with multiple constraints, verify that they are mutually compatible. If two constraints genuinely cannot coexist, decide which is higher priority and keep only that one.
Ask the AI if you need help understanding prompt constraints, how to write length or tone constraints, how to combine multiple constraints, or what to do when constraints conflict with each other.