Why Generic Tone Instructions Fail
Reviewed by Human · Updated August 5, 2026
'Just tell the AI to be professional and friendly.' It's the most repeated tone advice on the internet — and it's exactly why every AI-written email you've received this year sounds like the same beige robot wrote it. The model isn't ignoring your instruction. It's obeying it perfectly, and that's the problem.
✗ Wrong way
Write a welcome email for new users of our budgeting app. Make it professional and friendly.
You'll get something like: 'Dear Valued User, We're delighted to welcome you to our platform! We're committed to helping you achieve your financial goals...' Technically professional. Technically friendly. Also indistinguishable from ten thousand other welcome emails, because 'professional and friendly' describes ten thousand different voices — a law firm, a kindergarten newsletter, a bank, a barista. Asked to be all of them at once, the model produces their average.
✓ Better way
Write a welcome email for new users of our budgeting app. Audience: people who feel anxious opening their banking app. Sound like a calm friend who's good with money — think Oatly packaging copy, but for finance. Example of the voice: 'You did the hard part. You showed up. We'll handle the spreadsheet stuff.' Avoid: exclamation marks, 'we're delighted', and any sentence that could appear in a bank's email.
Every line closes off wrong interpretations. The audience descriptor ('people who feel anxious') sets the emotional register. The style reference (Oatly, but for finance) points at a recognizable voice instead of an abstract adjective. The example sentence shows the voice instead of describing it. The anti-example list bans the failure mode you've already seen. The model no longer has to guess which 'professional' you meant.
So why does 'like Oatly packaging, but for finance' work when 'professional and friendly' fails? The answer explains almost everything about tone control.
Why Does AI Ignore Your Tone Instructions?
It doesn't ignore them — it averages them. **Tone**: the attitude a piece of writing carries toward its subject and its reader. When a tone word maps to thousands of conflicting styles in the training data, the model can't pick one, so it blends them. The blend reads as 'generic AI voice.' The fix is never a better adjective; it's a narrower target. This is the same collapse trick behind role prompting — a persona is really a bundle of tone decisions with a name attached. Name the bundle precisely, and the voice snaps into focus.
Your turn
Rewrite this tone instruction so a stranger could not misread it: 'Write it in a professional but friendly tone.' Use at least one specific device — a style reference, an audience, or an example sentence.
Reflect
Could two different people read your rewrite and produce the same voice? That's the test 'professional but friendly' always fails.
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