AI Email Writer

Most emails are decided before they are opened. The subject line and preview text determine whether the message is read now, later, or never — and AI drafts are consistently weakest exactly there, defaulting to vague labels like "Quick question" or "Following up". Strong subject lines carry three things: the topic, the action wanted, and any deadline. "Q3 budget: approval needed by Friday" beats "Budget question" in every inbox on earth. If nothing is needed from the reader, say so — "FYI: vendor contract signed" lets people file it in two seconds, which they will thank you for. The first sentence should carry the ask, not the throat-clearing. Busy readers read the first line and the last line; everything between is skimmed. The professional pattern is context in one clause, request in the next: "We are finalizing the Q3 budget this week — could you approve the marketing line by Friday?" Background, justification, and options go below, for the minority who need them. When you generate a draft with this tool, write the topic field as the ask itself, not the theme. "Request 2-day deadline extension for the API migration" produces a sharper draft than "deadline extension" every time.

AI email tools are trained mostly on English correspondence, and it shows. Ask for a formal business email and you get an American formal email — which can read as cold, abrupt, or even rude in Seoul or Tokyo. Korean business email has structural conventions the models rarely apply unprompted. Recipients are organized as 수신 (to) and 참조 (cc), and the greeting names the person with title attached — 김민수 부장님, never a bare name. Titles (대리, 과장, 차장, 부장, 이사) function as the politeness architecture; omitting them reads as a status claim. The body opens with a relationship line ("안녕하세요, OO팀 OOO입니다") before any business, and requests are softened into capability questions: "검토해 주실 수 있을까요?" rather than "검토해 주세요." Japanese email is more codified still: お世話になっております as the near-mandatory opener, the company-then-name order, and a closing よろしくお願いいたします that has no real English equivalent. Western — especially US — norms invert this: directness signals respect for the reader's time. "I'm writing to ask for X by Friday" is polite precisely because it is efficient; padding it with three paragraphs of context reads as evasive. The practical move: tell the generator who the reader is culturally, not just their role. "Formal email to a Korean 부장 I have never met, written in Korean business style" produces a fundamentally different — and safer — draft than "formal email to a manager". And have a native colleague review anything high-stakes; models still miss honorific nuances that humans notice instantly.

"Make it friendly but professional" is the most common tone instruction and the least effective one. Every model has a house interpretation of friendly — usually an exclamation point and a "Hope you're doing well!" — that has nothing to do with your voice. Three techniques control tone far more reliably. First, constrain the surface features directly: sentence length, contractions, exclamation points, emoji, greeting formula. "Short sentences. Contractions fine. No exclamation points. No pleasantries beyond one greeting line" beats any adjective. Second, give a calibration sample. Paste two or three sentences you actually wrote and ask the tool to match their register. Models are dramatically better at imitating a sample than at interpreting a label — this is the single most underused feature of AI writing tools. Third, name the relationship and the stakes, because tone is a function of both: a vendor you may drop next quarter and a client you are apologizing to for a second missed deadline need different drafts even for the same message type. The apology case has its own rule: state the failure plainly, own it without hedging ("we missed the deadline", not "the deadline was missed"), give the fix and the prevention, and skip the adjectives. Over-apologizing reads as weakness in Western business culture and as insincerity almost everywhere.

Weak: "Write a friendly but professional follow-up."

Strong:
"Write a follow-up to a client whose project we delayed.
Relationship: 3-year client, warm but formal.
Tone rules:
- Sentences under 20 words. Contractions OK.
- No exclamation points, no 'hope you're well'.
- Own the delay in the first sentence ('we', active voice).
- One concrete recovery date, one prevention step.
Match this register: 'Thanks for flagging this early.
We'll have the revised scope to you Tuesday.'"

Most follow-ups fail twice: they come too soon, and they say nothing new. "Just bumping this to the top of your inbox" asks the reader to do the same work again with added guilt attached. A defensible cadence for business requests: first follow-up after three business days, second after a week, then one final message that closes the loop. Internal asks can move faster; cold outreach should move slower. Each message must reduce the reader's cost of replying, not just repeat the ask. Shrink the request ("even a yes/no on the direction helps"), add the deadline's consequence ("we lock the vendor list Thursday"), or offer a default ("if I don't hear back, I'll proceed with option A") — the default-option close is the highest-response pattern in professional correspondence because silence becomes an answer. The final message should release the pressure honestly: "Closing this out for now — if it becomes relevant again, happy to pick it up." It preserves the relationship and, paradoxically, gets more replies than a fourth nudge. One cultural note: this cadence is calibrated for Western norms. In Korean and Japanese contexts, a phone call or messenger ping after the first unanswered email is often more appropriate than a chain of written reminders directed at a senior person.

The example below shows the same request before and after applying the principles above: ask in the subject, ask in the first line, one topic, a default option, and a scannable shape. Whatever the tool produces, a sixty-second check protects you from the failure modes specific to generated email. Verify every fact you did not type: AI drafts routinely invent meeting dates, misstate what was previously agreed, and fill gaps with plausible specifics — a fabricated "as we discussed on Tuesday" can quietly become a commitment. Check the names and honorifics letter by letter; a misspelled name undoes an otherwise perfect email. Delete the filler the model loves ("I hope this email finds you well", "I wanted to reach out") and any sentence that repeats the subject line. Make sure the draft contains exactly one question or request — generated drafts often smuggle in two or three, and multi-ask emails get partial answers. Finally, read it once aloud in the recipient's position: if any sentence would make you wince, rewrite that sentence yourself. The tool drafts; you remain the author of record, and the recipient will hold you — not the model — to every word.

Before:
Subject: Question
Hi, I hope you're doing well! I was wondering if you
might have had a chance to look at the proposal we
sent over, and also whether the timeline works, and
if we should maybe set up a call sometime?

After:
Subject: Proposal sign-off — OK to start May 12?
Hi Sarah — could you confirm the proposal by Thursday?
Scope and pricing are unchanged from our April 28 call.
If I don't hear back, I'll assume the May 12 start
date works and send the kickoff invite.
Full timeline attached if you want the detail.