PromptSharp
Issue #004 · Sat, Jun 27, 2026
Sharpen how you instruct AI — from your first prompt to full agents. |
Persona and tone: control the voice, not just the words
Same facts, wrong voice = wrong output. Tone is a parameter — set it on purpose. |
Models default to a bland, hedging, mid-corporate voice. To override it, don't say 'be friendly' (too vague) — anchor the persona to a concrete reference and give tone rules: who they are, what they sound like, what they never do. Naming a recognizable style ('like a calm ER doctor explaining a result') compresses a hundred adjectives into one vivid target the model can actually hit. |
▸ Copy-paste prompt
You are writing a status-update email to customers after an outage.
VOICE: Calm, direct, accountable — like an experienced incident commander, not a PR department. Plain words. Short sentences.
RULES:
- Lead with what happened and that it's resolved. No 'we apologize for any inconvenience' clichés.
- Take ownership. No passive voice that hides who did what ('mistakes were made').
- One concrete thing we're doing so it won't recur.
- Max 120 words.
FACTS:
- 47-minute API outage at 09:12 UTC
- Cause: a bad config deploy
- Fix: rolled back; adding a pre-deploy check
Write the email. |
▸ Level it up
New to this: Instead of 'be professional', say 'sound like [a specific person/role]'. A concrete reference voice beats vague adjectives.
Vibe coder: Give 2-3 explicit 'never do this' tone rules (no clichés, no passive voice). Banning anti-patterns shapes voice faster than describing the target.
Advanced: Codify a brand voice as a reusable system block: persona + 5 do's + 5 never's + 2 before/after rewrites. Pin it as the system prompt for every customer-facing generation so tone is consistent across your whole agent fleet. |
Today's takeaway: Anchor tone to a concrete voice and ban the clichés — vague adjectives produce vague output. |
|
|
|
Know someone who'd sharpen their prompts with this?
Forward it. Refer 3 friends and we'll send you our private prompt-pack.
Refer & get the prompt-pack → |
|
|
|
|