Humanetext

Why Your Work Emails Sound Like AI

Corporate email conventions produce exactly the patterns people now read as machine-written. Which habits to drop, and what to write instead.

Writing craftBy Humanetext Editorial5 min read

The short answer

Work emails sound machine-written because office convention teaches the same habits models overproduce: hedging so nothing can be quoted against you, the passive voice so nobody is named, abstraction instead of specifics, and formulaic openings and sign-offs. Cut the opening pleasantry, name who does what by when, and delete the hedges on things you actually know. Most emails lose a third of their length and gain all of their clarity.

A colleague forwards you a message and says, half-joking, that it reads like ChatGPT wrote it. Nobody did. It was written by a person, at a desk, on a Tuesday.

The reason is that corporate email conventions and machine-generated text arrive at the same destination from different directions. Both are shaped to give offence to nobody and commit to as little as possible, and that produces a recognisable texture.

Which means the fix has nothing to do with AI. It is the same fix that has always made email better.

Four habits, all learned at work

Hedging so nothing can be quoted against you. I just wanted to check whether it might be possible to potentially move the deadline slightly. Every softener is insurance against being pinned to a request. Stack six of them and the reader cannot tell whether you are asking for something or musing.

The passive voice so nobody is named. It was decided that the timeline would be adjusted. By whom? The construction exists to avoid saying. Sometimes that is diplomatic and deliberate; more often it is habit, and it leaves the reader unable to tell who is responsible for what.

Abstraction instead of specifics. We're experiencing some resourcing challenges around delivery. Compare: Two of the four engineers are on the other project until the 15th, so we'll miss the deadline by about a week. The second is longer and actually says something.

Formula at both ends. I hope this email finds you well and please don't hesitate to reach out are pure ritual. Nobody reads them. They exist because you have seen them a thousand times.

Every one of these is also something language models overproduce, because they were trained on enormous quantities of exactly this register. That is the whole coincidence: models learned corporate email, and corporate email now sounds like a model.

What to cut

The opening sentence. Almost always a pleasantry or a restatement of why you are writing. I hope you're well. I wanted to reach out about the Q3 report. Both sentences can go — the subject line already said what this is about. Start with what you need.

"I wanted to" and "I just wanted to." These frame a request as a wistful hope. I just wanted to ask if you could send the figures becomes Could you send the figures? The second is a request, which is what it always was.

"Please don't hesitate to." Say let me know if you have questions, or say nothing, because they will anyway.

Stacked hedges. One qualifier is sometimes honest. Might potentially be able to possibly is four, and it means you are afraid of the sentence.

"As per," "at your earliest convenience," "circling back," "touching base." Ritual phrases that carry no information and mark the writing as institutional rather than personal.

What to write instead

The useful shape for most work emails is three things: what you need, from whom, by when.

Before: I hope this email finds you well. I wanted to reach out regarding the quarterly figures. It would be great if we could potentially get sight of those at some point this week if that's possible at all, as there are some downstream dependencies that may be impacted. Please don't hesitate to let me know if you have any questions.

After: Could you send the Q3 figures by Thursday? Finance needs them to close the month, and they can't start until we deliver. Shout if that's not doable and I'll tell them Friday.

Seventy-one words to thirty-nine, and the second one can be acted on without rereading. It also names the reason, gives a real date, and offers a way out — none of which the first one manages despite being twice the length.

The general principles are the same ones that improve any prose: be specific, cut the hedging, and use verbs rather than abstract nouns. They are covered at length in how to write like a human.

Formal does not mean stiff

The most common objection is that this reads as too casual for work. It confuses two separate things.

Formality is how much distance the situation calls for. An email to a regulator needs more than one to your team, and that is judgement, not skill.

Register is whether a person is discernibly behind the writing. That is available at every level of formality. A legal notice can be direct and clear while remaining entirely formal. A chatty message can be evasive and stiff.

When someone says an email sounds robotic, they almost never mean it was too formal. They mean nobody seemed to be there. The fix is directness and specifics, not jokes. More on the distinction in how to write conversationally without sounding fake.

Where the hedging is right

Not every softener is filler, and stripping all of them produces its own problem — an inbox of terse demands.

Genuine uncertainty deserves marking. If you do not know whether the deadline can move, say so, and say why. That is information, not evasion.

Requests up the hierarchy carry more cushioning, and pretending otherwise is naive about how offices work. Would you have time to look at this before Thursday? is appropriate; four stacked qualifiers is not.

Bad news needs care. Delivering it in five blunt words is not clarity, it is carelessness. Being clear about what happened and what you are doing about it is the goal, not brevity for its own sake.

The test is whether the softener is doing real work or is reflex. Ask of each one: am I genuinely unsure here, or am I protecting myself from a sentence I could stand behind?

Read it aloud before sending

The single most effective check, and it takes ten seconds on a short email.

If you would not say the sentence to that person's face, do not send it. I hope this email finds you well fails immediately — nobody has ever said that out loud to a colleague. So does at your earliest convenience.

Reading aloud also catches the emails where you buried the request in paragraph three, because you will hear yourself get to the point too late.

The reason this matters beyond style

Emails that hedge everything and name nobody do not just read badly. They fail at their job.

The recipient cannot tell what you want, whether it is urgent, or whether they are the person who should act. So they skim it, file it, and nothing happens. Then you follow up, which is another email, and the loop continues.

Directness is not a stylistic preference here. It is the difference between an email that produces an outcome and one that produces another email.

Common questions

Why do my emails sound robotic?
Usually four learned habits rather than anything to do with AI: hedging every claim so it cannot be held against you, using the passive voice so no one is named, reaching for abstract nouns instead of specifics, and opening with a formula. Office writing rewards all four because they reduce personal exposure, and all four also happen to be what language models overproduce.
How do I make my emails sound more human?
Delete the first sentence, which is almost always a pleasantry or a restatement of why you are writing. Say who needs to do what by when, in the active voice. Use contractions. Cut hedges on anything you actually know. Then read it aloud — if you would not say it to their face, rewrite it.
Will my colleagues think I used AI to write this?
Increasingly people do wonder, and the irony is that the emails most often suspected are the polished formal ones rather than the casual ones. If you want to be recognisably yourself, be specific and direct: name people, give dates and numbers, and state your actual view. Generated text is general and hedged by default.
Is it bad to use AI to write work emails?
For routine correspondence, most people would not care and no norm treats it as disclosable. The risk is different: an AI-drafted email is generic by default, and generic messages get skimmed and forgotten. If the email matters, the specifics that make it land are the part you have to supply anyway.
What phrases should I stop using in emails?
I hope this email finds you well, I wanted to reach out, please don't hesitate to, as per our discussion, at your earliest convenience, and circling back. Each is filler that delays the point, and every one appears constantly in generated text because it is heavily represented in the corporate writing models trained on.
How long should a work email be?
Short enough that the recipient can act without rereading. For most requests that is three to five sentences: what you need, from whom, by when, and why it matters. Anything longer usually means the email is doing the work a document or a meeting should do.

Keep reading