Specificity: The Single Biggest Lever
Reviewed by Human · Updated July 13, 2026
Real world
Walk into a café and say 'coffee, please.' You'll get the house drip. It's fine. It's nobody's favorite. Now say 'a double oat-milk cortado, extra hot, to go' — and you get your drink, made your way, first try.
AI parallel
A vague prompt is 'coffee, please.' The model hands you the house drip: safe, average, fine. A specific prompt is the full order. Same barista, same machine — completely different drink. The model's ability never changed. Your order did.
Five dimensions turn a foggy prompt into a sharp one. One of them gets skipped in almost every beginner prompt. Guess which.
How Do You Make a Prompt More Specific?
Answer five questions before you hit send. **Audience**: The specific person or group the output is written for. **Format**: The shape of the output — email, table, bullet list, script. **Length**: The size of the output, stated in words, sentences, or items. **Tone**: The voice of the output — formal, casual, urgent, playful. **Purpose**: The job the output must do once it exists — persuade, inform, reassure, decide. Audience and purpose are context, and context is everything to a model deciding what to write. One more path exists: instead of telling the model what you want, you can show it with examples — that's few-shot prompting, and it gets its own course.
The same request, before and after specificity
✗ The foggy version
Which team? Which deadline? Moved up or pushed back? Is this good news or bad? Should they do anything? The model guesses all five — and writes a bland memo that answers none of them.
✓ The sharp version
Audience (six-person engineering team), length (120 words), context (the deadline moved, and in which direction), tone (reassuring), and purpose (get estimates updated) are all decided. The model executes instead of guessing.
Your turn
Take this prompt: 'Tell me about marketing.' Fill in all five dimensions, then write the rewritten prompt as one clean paragraph.
Reflect
Notice how deciding the five dimensions forced you to figure out what you actually wanted. That's not a side effect. That's the point.
Field note
Early on, a client asked me to draft the announcement for their app's biggest feature launch of the year. My prompt was basically: 'Write an announcement post for a new expense-tracking feature. Professional but friendly.' The draft came back clean, polished — and the client rejected it in one line: 'This reads like it's for accountants. Our users are freelancers who hate accounting.' That stung, because the model hadn't failed. I had. I'd never told it who was reading. I changed one thing — the prompt became: 'Write an announcement post for a new expense-tracking feature. The readers are freelancers who dread bookkeeping and put it off until tax season. Lead with the relief, not the feature.' Same model, same day, same feature. The new draft opened with 'You know that shoebox of receipts you're avoiding?' — and the client approved it without edits. One sentence of audience did what three rounds of tone tweaking couldn't. I've never sent a client-facing prompt without an audience line since.
Constraints make AI more creative, not less
Counterintuitive but true in practice: a wide-open prompt pulls the model toward the most statistically common answer — the cliché. Tight constraints ('a bakery name, two words, no puns, must reference fire') push it off the beaten path, because the beaten path no longer fits the rules. Fence the field, and the model stops grazing in the middle of it.
You've finished this module.
Mark it complete to earn your XP and keep your streak alive.
Progress saved locally · Sign up to earn XP