AI for Everyday Work

Lesson 4 of 14

Rewriting and tone

Here is a secret that surprises people: AI assistants are better at rewriting than at writing. Given a blank page they produce competent filler; given your rough draft they produce your draft, but better — and rewriting is where the emotional labour of office communication actually lives. The email you rewrote three times, the message you softened and re-softened, the paragraph you know is too long but can't see how to cut: this lesson is about handing that specific misery to the machine.

Priya's vendor email

Remember the email Priya drafted twice and deleted twice. Her honest draft:

text
This is the third time this quarter your trucks have arrived late and
honestly it's becoming impossible to defend you internally. We had a
client escalation because of Tuesday's delay. If this happens again
we will have to look at other vendors. Fix it.

Every word is true, and she can't send it — the vendor relationship has to survive the message. The ask:

text
Rewrite this so it's firm but professional. Keep the facts and the
seriousness — I want them to feel the relationship is genuinely at
risk — but remove the anger. Keep it short. Keep my plain style,
don't make it corporate.

What comes back keeps the spine — three incidents, the client escalation, the explicit warning — wearing language she can stand behind in a meeting with their CEO. Thirty seconds, and the version she sends is both more professional and more effective than the angry one, because the recipient reads the whole thing instead of getting defensive at line one.

The register moves

Tone is one dial; there are several, and they all work the same way — state the direction, paste the text:

  • Shorter — "cut this by half without losing the commitments." The model is a ruthless, egoless editor of your darlings.
  • Simpler — "rewrite so someone outside logistics follows it." Jargon removal for the client-facing version.
  • More formal / less formal — up for the regulator, down for the team channel.
  • Different reader, same facts — "make this an update for the COO" vs "make this a heads-up for my team": emphasis shifts, facts stay.
  • Structure change — paragraphs into bullets, bullets into prose, email into a one-line summary plus details.

The pattern in all of them: the content is yours and already exists; the model is only transforming its surface. That is why this is the most reliable zone on the whole capability map — there is nothing to hallucinate.

Keeping your voice

The one real risk of habitual rewriting is coming out the other side sounding like everyone else — AI's default register is smooth, agreeable, and slightly beige. Two defences. First, ask for it: "keep my phrasing wherever possible; change only what the tone requires" produces a light-touch edit instead of a full repaint. Second, always do a final pass yourself and put back one or two phrases that are unmistakably you. Priya's closing line — "I'd rather fix this with you than start over with someone new" — was hers, not the model's, and it is the best line in the email.

A worthwhile companion habit: when a message really matters, ask "how might the recipient take this badly?" before sending. The model is genuinely good at spotting how a sentence lands wrong — it has read more misunderstood emails than anyone alive.

Do this today: find the last message you hesitated over before sending, run the firm-but-professional pass on it, and compare. Then try the same message aimed at a completely different reader, just to feel how cheap the transformation is.

← Previous