How to Use AI for Emails Without Sounding Like a Robot Wrote Them

An email is not a small essay. It has its own genre rules, and most of the advice about AI writing in general skips right past the specific things that give an AI-drafted email away.

Most writing advice, including a fair amount of what we've published here before, treats "AI writing" as one big undifferentiated category with the same fixes everywhere: vary sentence length, cut hedge phrases, add specific detail. That advice is real and it works, but email has its own extra layer of genre conventions on top of it, things that are fine or even expected in a blog post and read as immediately wrong in someone's inbox. A subject line isn't a headline. A sign-off isn't a conclusion. Email has a shape all its own, and getting that shape wrong is often a faster tell than any individual sentence.

The subject line problem

AI-generated subject lines tend toward two failure modes: too vague to be useful, "Quick Update" or "Following Up," or too formal for the actual content, a full sentence with a period where three words would do. A good subject line front-loads the one thing the recipient actually needs to know before they open it, the specific ask, the specific update, the specific deadline. "Need your sign-off by Thursday" does more work than "Requesting Your Review and Approval."

The opening line

"I hope this email finds you well" is probably the single most recognizable piece of AI-generated email boilerplate in circulation right now, and it was already tired before AI started writing it constantly. It delays the actual point of the email by a full sentence for no benefit, the recipient already knows you're about to ask them something or tell them something, that's why they're reading. Getting to the point in the first sentence isn't rude, in most professional contexts it's read as respecting the other person's time, which it actually is.

This doesn't mean every email should open cold with zero warmth, especially with someone you know. It means the warmth, if there's room for it, should be specific rather than a formula. "Hope the move went okay last weekend" says you actually remember something about this person. "I hope this email finds you well" says nothing about anyone.

The sign-off

"Please don't hesitate to reach out if you have any questions" is doing the same job as the "I hope this finds you well" opener, it's a formula standing in for an actual thought. If there's a real next step, say what it is. "Let me know by Wednesday if this works" gives the recipient something concrete to act on. If there's really nothing left to say, a short closing is fine, it doesn't need padding to feel complete.

Matching register to relationship

The right sign-off for a cold outreach email to someone you've never met is not the right sign-off for a quick note to a coworker you talk to daily, and AI drafts often default to one formal register regardless of who's actually receiving the email. A model that doesn't know your relationship to the recipient will guess at something safely formal, which is exactly why it's worth telling it, briefly, who this is going to and how well you know them, before asking for a draft.

Length calibration

AI drafts tend to pad short emails with unnecessary context and structure, three paragraphs where two sentences would do the job, because the model doesn't know how much the recipient already knows or how much time they have. A quick internal note asking someone to confirm a meeting time doesn't need a background paragraph. A proposal to a new client might actually need more context. Tell the model which situation you're in, or better, just cut the draft down yourself once you have it, most AI-generated emails can lose a third of their length without losing anything the reader actually needed, and a shorter email is more likely to actually get read in full rather than skimmed and half-answered.

The "just circling back" problem

Follow-up emails are their own particular minefield. AI drafts default to a small set of phrases for these, "just circling back on this," "wanted to bump this to the top of your inbox," "following up in case this got buried," and all of them have become so common that they've started to feel slightly passive-aggressive even when no aggression was intended. Part of the problem is that these phrases apologize for the follow-up rather than just doing it. A direct restatement of the actual ask, with a specific new deadline or piece of context if one exists, reads as more confident and, frankly, more respectful of the recipient's time than a soft preamble about circling back.

Worth noting the opposite failure too: some AI drafts overcorrect into being oddly blunt in a follow-up, which can read as curt with someone you don't know well. There's no formula that works in either direction here, the actual fix is naming the real situation. If a deadline moved, say so. If you're following up because it's actually time-sensitive, say that specifically instead of leaving it implied.

Internal notes versus external mail

Email conventions have quietly split into two different genres over the last decade, without most people noticing it happened. A quick internal note to a coworker has drifted much closer to a chat message in tone, shorter sentences, less formal structure, sometimes no greeting at all beyond a name. An external email to a client, a vendor, or someone outside the company has held onto more of the traditional structure, a real greeting, a clear close, complete sentences throughout. AI models trained broadly tend to default toward the more formal version regardless of which situation you're actually in, because formal is the safer statistical guess when the model doesn't know the context.

This is worth being explicit about in a prompt rather than assuming the model will infer it. "This is an internal note to a teammate I talk to daily" produces a noticeably different draft than "this is a first email to a potential client." The gap between those two isn't only politeness level, it also shows up in sentence length, greeting style, and how much context gets spelled out versus assumed. Getting this one setting right upfront saves more editing than almost any other single instruction.

A quick workflow

1. Tell it who's reading this and how well you know them. That one piece of context changes the appropriate register more than almost anything else.
2. Ask for the point in the first sentence. Explicitly, if needed, "start with the actual ask, no warmup."
3. Cut the formula phrases on sight. "I hope this finds you well," "please don't hesitate to reach out," "I wanted to reach out because." All three can usually be deleted with nothing lost.
4. Read it as if you're the recipient, on a phone, with thirty seconds. If the actual ask isn't obvious in that scan, move it earlier.

The actual point of an email

An email exists to get something across quickly and get a response, not to demonstrate that a complete, well-structured piece of writing was produced. That's a different goal from most of the writing this blog usually covers, and it's worth remembering when a draft comes back from a model looking polished but padded. Polished isn't the target. Fast to read, clear about what's needed, and appropriately warm for who's on the other end, that's the actual target, and it's a target a model can hit reliably once it knows enough about the situation to aim at it properly.

None of this requires a lot of extra effort once it becomes a habit. Most of the fixes here amount to telling the model two or three things it can't otherwise guess, who's reading this, how well you know them, and what the actual point is, before asking for a draft, then trimming whatever formula language slipped in anyway. That's a smaller ask than it sounds, and it turns a generic template into something that actually reads like a message from one specific person to another, which is really all a good email has ever needed to do. The same principle shows up again in how AI handles meeting notes, where knowing what the output is for matters as much as anything about tone.