writing with AI, speaking without it: keeping your own voice

Three people, one tool

A product manager in Bengaluru uses an AI assistant for every client email. The emails are flawless. Then a client calls her directly, asks a sharp question, and she hears herself hesitate in a way her emails never do.

A founder in Kochi posts on LinkedIn twice a week with AI help. His reach is fine. A friend tells him, kindly, that the posts "sound like everyone else's" and that she misses the way he actually talks.

A team lead in Dubai has to tell a colleague that a project slipped because of his work. She asks an AI tool to write the message. It is balanced, polite and long. The colleague replies with one line: "Is this from you?"

All three are using a good tool. The problem is not the tool. It is what each of them handed over to it.

What AI is good at, and where it slips

AI drafting tools are strong at a first version, a tidy structure, a summary of a long thread, and catching grammar slips. For many professionals writing in a second language, they remove hours of worry each week.

They are less reliable in ways that matter. Researchers from Harvard Business School, Wharton, MIT and Warwick ran a field experiment with Boston Consulting Group consultants and found that AI helped a great deal on some tasks and made results worse on others that looked just as hard. They called this a "jagged technological frontier". The practical lesson for anyone writing at work: the tool does not know which side of the line your task is on. You do.

So treat the draft as a draft. You are still the author, and the person who will be asked about it on the next call.

Editing AI text back into your voice

AI writing has tells that readers now notice quickly: an opener that praises the question, three adjectives where one would do, lists of exactly three, and endings that offer more help. A few edits bring your voice back.

• Cut the first and last sentence. They are usually throat-clearing.
• Add one detail only you could know. A name, a date, what happened on Tuesday's call.
• Replace big words with your words. If you would never say "leverage" aloud, do not send it.
• Shorten. Most AI drafts are a third longer than they need to be.
• Read it aloud. If it does not sound like you speaking, change it until it does.

When not to use it at all

Some messages carry their value in the fact that you wrote them. When AI writes them, people can often tell, and the message loses the very thing it was meant to give.

• Apologies. An apology works because it costs you something. Say what you did, that you are sorry, and what you will do now.
• Condolences. Two plain sentences of your own mean more than a beautiful paragraph from a tool.
• Feedback. The person needs to know it is your view, in your words, and that you will stand behind it in conversation.
• Thanks. Name the specific thing they did.
• Anything personal. A message to a parent, a partner, a friend who is struggling.

"I'm so sorry about your father. I remember him at your wedding, telling everyone about his garden. I'm thinking of you."

That message needs no editing and no tool.

The gap between your messages and your voice

Here is the pattern coaches now see often. Someone's emails are clear, structured and confident. On a live call, the same person rambles, apologises, or goes quiet. Colleagues notice the difference, even if they never mention it.

The gap matters more as AI spreads. When every email is fluent, written polish stops telling people much about you. What still shows who you are is the live moment: a question you did not expect, a client call, the review with your manager, a town hall. You cannot paste a reply into a conversation that is happening now. This is our view, not a measured finding. It sits comfortably with the World Economic Forum's Future of Jobs Report 2025, in which employers place human skills such as analytical thinking, resilience and flexibility, and leadership and social influence among the skills they treat as core, alongside fast-growing technology skills.

The fix is simple to describe: practise saying out loud what you now only write.

"The short version is this: we're two weeks behind because the vendor changed the spec. I've got two options for catching up, and I'd like your view on which one."

If you can say your key email in two spoken sentences, you are ready for the call that follows it.

Using AI to rehearse

AI is a useful sparring partner for speaking, as long as you keep the final performance for yourself.

• Role-play the other person. Ask a voice or chat tool to play a client who pushes back on price, or a senior who says no.
• Ask for the hardest questions. "What are five tough questions a CFO might ask about this plan?"
• Answer aloud, not in text. Typing your answer trains writing. Speaking trains speaking.
• Then put it away. Do the real conversation from three notes, not a script.

"Play my manager. I'm asking to move my review to next month because the quarter-end close overlaps. Push back twice before you agree."

Sources

1. Fabrizio Dell'Acqua, Edward McFowland III, Ethan Mollick and colleagues, Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of AI on Knowledge Worker Productivity and Quality, Harvard Business School Working Paper 24-013, 2023. www.hbs.edu/faculty/Pages/item.aspx?num=64700
2. World Economic Forum, The Future of Jobs Report 2025, 2025. www.weforum.org/publications/the-future-of-jobs-report-2025/

Published by The Full Stop, the journal of Brightside English. Reviewed by a Brightside senior coach. Corrections: see our editorial standard.