Addaly is in open beta. Things will change, and AI answers can be wrong — check anything that matters.

Drafting, rewriting, and the tone problem

AI at Work · lesson 2 of 8 · 9 min

You can get a usable draft out of AI in one try. Most people do not, and they blame the tool. The gap is almost always the same: they asked for a *thing* instead of describing a *situation*.

The four things a draft needs

A colleague who writes well for you knows four things you rarely bother to say out loud:

  1. 1Who is reading it, specifically. Not "a customer" — "a landlord of 30 years who is angry and has already emailed twice."
  2. 2What you want to happen next. Not "inform them" — "get them to send the meter reading by Friday without making them feel accused."
  3. 3The facts. Names, numbers, dates, what actually happened.
  4. 4Constraints. Length, language, tone, format, what must not be said.

Compare. Weak:

Write an email to a client about a delayed delivery.

Strong:

A client in Lagos ordered 200 printed folders for a conference on 14 March. Our printer's machine failed; we can deliver 200 folders on 12 March but only in matte, not the gloss they chose. They have been a customer for two years and are usually easy-going. Write an email, under 150 words, that leads with the solution rather than the apology, offers a 10% credit, and does not use the phrase "unfortunately".

The second prompt has three sentences of context and produces something you might actually send. That ratio holds almost everywhere: context in, quality out.

Rewriting is stronger than generating

The highest-value move is not "write me an email." It is "here is my email — make it work." You supply the judgment and the facts; the model supplies the sentence-level craft.

Useful rewrite instructions, all of which do specific work:

  • "Cut this by 40% without losing any number or date."
  • "Rewrite for someone who left school at 15."
  • "Make this firmer. Right now it sounds like I am apologising for asking to be paid."
  • "Same content, but as five bullet points a busy person can read on a phone."
  • "Keep my wording wherever you can. Only fix what is unclear."

That last one matters more than it looks. Left alone, models flatten your voice into the same beige corporate register — the em-dashes, the "delve", the "it's not just X, it's Y", the three-item lists everywhere. Readers are learning to spot it, and it reads as *nobody bothered*.

Getting your own voice back

Paste in two or three things you have genuinely written — a past email, a note to your team — and say: match this voice. Short sentences if you write short sentences. Contractions if you use them. This works better than any adjective you could pick.

Then, always, do a last pass by hand. Delete the throat-clearing opener ("I hope this email finds you well"). Cut any sentence that says nothing. Put back the one blunt line you would actually have written. Thirty seconds of this is the difference between a draft that sounds like you and one that sounds like it came from a machine, because the second kind quietly tells your reader how little they were worth.

One warning

Asking for a draft "in the style of" a real named colleague or a specific public figure, and then sending it as if it were theirs, is not a style choice. It is impersonation. Style-matching your own writing is fine. Style-matching someone else's signature is not.

Before you move on

A school administrator gets stiff, generic letters from AI no matter what she tries. She has been prompting: "Write a warm, friendly, professional letter to parents about the new pickup times." What is the most likely reason the tone still feels wrong?

Pick the one you would defend. Nobody sees your answer.

No ads. No data sale. No public scores on people. Ever.

© 2026 Addaly