what AI-written emails do to how we speak out loud

The voice that arrives first

Picture a procurement manager in Riyadh. For three weeks she has been exchanging emails with a supplier's account lead in Chennai. His emails are excellent: crisp, warm, perfectly structured, with the right amount of formality. She has formed a picture of him. Confident, senior, quick.

Then they speak on a call. He is thoughtful and capable, but he is also careful, a little slower, and he says "sorry" twice in the first minute. Nothing is wrong. He is simply not the man in the emails.

She does not think less of him. But she adjusts. And he feels her adjusting, which is worse.

We use this illustration because the pattern behind it is one our coaches meet often now. The written version of a person has started to run ahead of the spoken one. For a long time, our writing and our speaking were roughly the same person on two different days. Now one of them has an assistant.

Borrowed confidence

Here is the uncomfortable part. A polished AI draft does not only fix grammar. It changes tone. It removes hesitation. It adds certainty that the writer may not feel. "We are confident this approach will deliver" sounds lovely on the page. Would you say it, in that exact way, to a sceptical client looking at you?

When we send words like that, we are borrowing confidence. Borrowing is not a crime. The problem is that the loan is called in the moment we open our mouths. The reader has met the borrowed version, and now expects it in person.

This is why, in our view, AI-written emails can quietly raise the stakes of speaking. They set an expectation that the live voice has to meet. For someone who already finds spoken English harder than written English, that is an extra weight carried into every call.

Who owns a sentence?

We have a simple working test for ownership, and it has nothing to do with who typed it.

A sentence is yours if you can say it out loud, in your own rhythm, and defend it when someone pushes back.

By that test, plenty of AI-assisted sentences are fully owned. You asked for a draft, you read it, you changed the parts that were not you, and you could explain every claim on a call tomorrow. Good. The tool did what a thoughtful colleague might do.

But some sentences fail the test. The writer would not use that phrase aloud. The writer could not explain what "a holistic, scalable framework" actually means in this project. The writer would be uneasy if the reader replied, "Tell me more about that." These sentences are rented. They look like ours in the inbox, and they abandon us in the conversation.

Ownership matters because communication at work is not a series of documents. It is a relationship. Every email is a promise that the person behind it will show up and stand by it.

What the page cannot carry

A polished page is wonderful at some things: structure, accuracy, a record. But the live voice carries things the page cannot.

It carries pace, which tells the listener how sure you are. It carries the small laugh that says you are human. It carries the pause before a hard sentence, which tells someone you know this is hard. It carries your accent, which tells them where you are from and that you are not pretending otherwise. And it carries your reaction to them, in real time, which no draft can predict.

When the page and the voice disagree, people tend to trust the voice. That is not a rule we can cite. It is what we see. The client remembers how you sounded when the price came up, not how elegantly you phrased the proposal.

The quiet effect on speaking

There is another effect, slower and less obvious. If we hand over our writing completely, we stop rehearsing.

Writing an email yourself used to be a small rehearsal for saying it. You searched for the phrase, you weighed two options, you chose. That choosing is the same muscle we use when we speak. A professional who writes thirty of her own emails a week is, without knowing it, practising how to put her thoughts into English thirty times. A professional who pastes thirty AI drafts a week may be practising something else: approving.

Approving is a real skill. It is just not speaking.

We are not arguing against the tools. We use them. We are arguing that if the tools now do some of our rehearsal, we need to put that rehearsal back somewhere else, deliberately, out loud.

Making borrowed words your own

There is a hopeful side to all this. An AI draft can be a source of new language, in the same way a good book or a well-spoken colleague is. The trick is to do what learners have always done with good language: take it, try it in your mouth, keep what fits, discard what does not.

Our coaches suggest a simple habit. When an AI draft gives you a phrase you like, say it aloud three times in your own voice. If it feels natural by the third time, keep it, and use it on your next call. If it still feels like a costume, cut it from the email too. The page should never use words the voice would refuse.

Over time, the gap narrows from both ends. Your emails become a little more like you, and your voice picks up a few of the phrases that genuinely help.

The honest email

We will end with a small argument for imperfection.

An email that sounds like the person who wrote it, slightly plainer, a little less shiny, prepares the reader for the real conversation. The manager in Riyadh, if she had received emails in the account lead's own voice, would have met the same thoughtful, careful man on the call that she had met in the inbox. No adjustment. No gap. Just trust building in a straight line.

That, in the end, is what we want from our words. Not to impress once, but to be believed every time.

Published by The Full Stop, the journal of Brightside English, in the Thought Studies series. This is an essay: an argument and an experiment of the imagination, not research. Reviewed by a Brightside senior coach. Corrections: see our editorial standard.