Add constraints

Constraints are the boundaries of an answer: how long, what is in scope, what must survive, what must not appear. They are the difference between a good draft and a usable one — and they double as the checklist you run when the result comes back.

The four families

1. Length — words, bullets, sentences, table rows. Numbers, not adjectives: “120 words”, not “short”.
2. Scope — what is included and excluded: “only the numbers below”, “no features outside this list”.
3. Must-keep — facts, commitments, phrases that cannot be lost: “keep every deadline”.
4. Must-avoid — the classic failure modes: “no new facts”, “no client names”, “no promises”.

Two or three constraints usually do the real work. More than five and they start fighting each other — when that happens, say which wins: “if tone conflicts with length, keep the length.”

Why vague constraints fail

“Short” means 20 words to one writer and 200 to another. “Don’t be boring” names something to avoid without naming the target — and models follow positive direction far better than negative vibes. Give the boundary a number or a shape, and keep the don’ts for things that are actually dangerous.

A bad example

Ask: “Write a product update. Keep it short and professional and don’t oversell.”

Which parts are overselling? Short to whom? You get a plausible middle-length update that violates nothing specific and fits nowhere.

A better example

Ask: “Write a product update for existing customers: maximum 120 words, three short paragraphs (what changed, what it means for them, what to do next). No marketing adjectives; no client names; no promises about future releases.”

Every failing draft now fails specifically — one paragraph too many, an adjective, an invented promise — and each fix is one edit away.

Why it works

Constraints narrow the space of plausible tokens — the same mechanism as everything else in prompting, applied to the edges. The bonus is diagnostic power: with stated boundaries, review stops being taste and becomes a checklist. That is also what makes prompts reusable: the boundaries are the brief.

Constraints also make results comparable: same length, same sections, same tone across ten options means you choose on substance instead of presentation. That is why constraint-heavy prompts pay twice — once on the first draft, and again every time you reuse the prompt for the next item in the queue.

Practice

Transfer

• Emails: cap the sentences; forbid new promises.
• Reports: scope to the listed data; require the caveats.
• Anything a human will skim: one number for length, one rule for tone.

Next

Next: Specify the format — choosing the shape of the answer before you get it. The prompt glossary entry covers the basics.