Chapter 1 · Lesson 2
Set Constraints That Improve the Work
Use limits that guide quality instead of adding arbitrary prompt decoration.
Your AI drafts are polished but too long, too vague, and full of claims your team cannot support.
A constraint should protect the outcome
Good constraints reflect a real need: available time, approved evidence, reading level, brand voice, legal boundaries, or the decision-maker’s attention. “Use exactly 317 words” is rarely useful. “Keep it under 200 words so it fits the weekly briefing” explains why the limit exists.
Tell AI what must not happen
Negative boundaries can be valuable when they are specific. Ask the model not to invent figures, not to add claims beyond the supplied sources, or not to change agreed terminology. Avoid long lists of vague prohibitions that compete with the actual task.
Prioritise when constraints conflict
Models often try to satisfy every instruction even when the instructions pull in different directions. State the priority: accuracy before brevity, clarity before warmth, or source fidelity before creativity. This gives the model a sensible trade-off when it cannot maximise everything.
Build the mental model
Constraints improve work when they express real trade-offs and make priorities visible.
Separate hard rules from preferences
A hard rule must be satisfied, such as using approved sources. A preference can bend, such as keeping the answer near 150 words. Label them so the model knows what to protect.
Give a conflict rule
When requirements compete, state the order: source accuracy first, required content second, brevity third. This prevents the model from hiding omissions to meet a word limit.
A customer update must be brief without inventing certainty.
Keep it short, friendly, accurate, complete, confident, and detailed.
Hard rules: use only confirmed facts; label unresolved dates as pending; include the customer action. Preference: stay under 180 words. If brevity conflicts with accuracy or the required action, keep the accurate detail and exceed the target slightly.
The improved version distinguishes non-negotiable rules from a flexible target and explains how to resolve the likely conflict.
A quick constraint audit
For each constraint, ask what failure it prevents. If you cannot name the failure, remove or rewrite the constraint.
Check the set for contradictions before prompting. “Complete” and “under 50 words” may be incompatible; “confident” and “state uncertainty” need a priority rule.
Turn vague quality words into operating rules
Take a prompt that says clear, accurate, or professional. Replace each adjective with an observable requirement.
Hard rules: [...] Preferences: [...] If they conflict, prioritise [...] because [...].multiple choice
Which constraint is most likely to improve a customer update?
Choose the answer that best fits the principle from this lesson.
Useful constraints come from the real conditions surrounding the work.