LESSON 1 OF 5
Constraints beat length
BY THE END OF THIS LESSON
Improve a prompt by removing options rather than adding words.
Long prompts are not better prompts
Once people learn that detail helps, they write essays. But a 600-word prompt full of hedging and background is often worse than a 90-word one, because it buries the actual requirement.
What improves output is not word count. It is how many wrong answers your prompt rules out.
Think in terms of the answer space
Every prompt defines a set of answers that would satisfy it. "Write about productivity" admits millions. "Write 150 words for an audience of hospital shift managers, arguing against morning stand-ups, with one concrete alternative" admits very few.
Each constraint you add cuts the space. The craft is choosing constraints that cut away the answers you do not want while leaving the ones you do.
The four constraints that do the most work
Audience — who is reading, and what they already know.
Length — a number, not "concise". Concise means nothing; 120 words means something.
Exclusions — what it must not do. Often the single highest-leverage line in a prompt.
Success criteria — how you will judge it. "It works if a busy reader gets the point from the first sentence alone."
WORKED EXAMPLE
LONG BUT LOOSE
I need some help writing about our new feature. We are a software company and we have been working on this for a while and it is quite exciting. It lets people schedule things more easily. Please write something good that would appeal to our customers and make them want to use it. Make it engaging and professional but not too formal. Thanks!
SHORT BUT TIGHT
Audience: operations managers at mid-size warehouses who currently schedule shifts in a spreadsheet. Write 120 words announcing auto-scheduling. Must include: the specific pain (rebuilding the rota every time someone calls in sick). Must not include: the words "excited", "revolutionary", or any mention of AI. It works if a manager reads the first sentence and recognises their own Monday morning.
The first prompt is longer and says almost nothing enforceable — "good", "engaging", "not too formal" are all unfalsifiable. The second is shorter and every line rules something out.
YOUR TURN
Have the model find the loose parts of your own prompt.
RUN THIS
Here is a prompt I use. Do not answer it — audit it. For each instruction, tell me whether it actually rules any answers out, or whether it is unfalsifiable filler like "make it engaging". Then give me the tightest version that keeps every real constraint and deletes the rest. My prompt: [paste yours here]
What a good result looks like
Expect the rewrite to be shorter than your original. If it is longer, reply "cut it by half without losing a real constraint".
KNOWLEDGE CHECK
Answer all 3 correctly to complete this lesson.
1. What actually makes a prompt more effective?
2. Which instruction is unfalsifiable filler?
3. Why is an exclusion list often the highest-leverage line in a prompt?
REMEMBER THIS
A better prompt rules out more wrong answers — that is usually shorter, not longer.