How to Use AI at Work Without Sounding Robotic

Generative AI has become an everyday tool for many working professionals. People use it to draft emails, summarise meetings, tidy reports and prepare for conversations they would rather not have. The time saved is real. The difficulty is what happens to the messages. When the first draft is sent as it is, emails start to sound alike: polite, long, slightly formal and somehow empty, and readers often notice.

This article covers how to get the productivity benefit of AI, including simple automation, without handing over the part of communication that builds trust.

Why “just use ChatGPT” saves time but can cost trust

The productivity case is well supported. In an experiment published in Science, Shakked Noy and Whitney Zhang set 453 college-educated professionals writing tasks from their own occupations and randomly gave half of them access to ChatGPT. The average time taken fell by 40% and output quality rose by 18%.

A larger field study points the same way. When a generative AI assistant was introduced to 5,172 customer-support agents, Erik Brynjolfsson, Danielle Li and Lindsey Raymond found that issues resolved per hour rose by 15% on average. Less experienced agents improved in both speed and quality. Customers were also more polite and less likely to ask to speak to a manager. Both studies looked at specific writing and support tasks rather than every kind of work, but the direction is clear.

The cost shows up elsewhere. In two randomised experiments published in Scientific Reports, Hohenstein and colleagues found that AI-suggested replies made conversations faster and more positive, and partners rated each other as closer and more cooperative. However, “people are evaluated more negatively if they are suspected to be using algorithmic responses.” The participants were online crowdworkers rather than Malaysian colleagues or customers, but the lesson carries: AI helps until the reader suspects it. That makes sounding robotic a trust problem, not only a style problem.

Decide which messages AI drafts and which you write yourself

Routine, information-heavy writing is a good fit for AI. That includes meeting summaries, first drafts of reports, internal updates, standard replies to common enquiries, and turning rough notes into clear English or Bahasa Malaysia.

Messages where the reader needs to feel that a person thought about them are different. Apologies to a customer, condolences, difficult feedback to a team member, and replies to a long-standing client who knows how you write should start with you.

A widely reported case shows what happens when that line is crossed. In February 2023, an office at Vanderbilt University’s Peabody College emailed students after a shooting at another university. The email ended with a note that it was a “paraphrase from OpenAI’s ChatGPT AI language model”. The college apologised, saying that “using ChatGPT to generate communications on behalf of our community in a time of sorrow and in response to a tragedy contradicts the values that characterize Peabody College.” It said it still believed in the message itself. What it misjudged was that readers expected to hear from a person.

Give AI your context before you ask it to write

Robotic drafts often come from thin instructions. “Write a follow-up email to the client” gives the AI nothing to work with, so it fills the gap with generic phrases. Before asking for a draft, tell it:

• Who the reader is and how well you know them
• How you normally address them (first name, Encik, Puan, Dato’) and which language you usually use together
• The one thing you want them to do after reading
• Two or three facts only you know, such as what was agreed in the meeting, the figure or the deadline
• A short sample of a message you wrote yourself, so it can match your length and tone

Edit the draft until it sounds like you

Treat every AI draft as a first draft. Before sending, check:

• Does the first line say why you are writing? Cut generic openings and get to the point.
• Is it the length you would normally write? AI drafts often run longer than a person would.
• Does it include specifics only you would know? Replace general statements with the actual date, figure or decision.
• Would you say this to the person’s face? Read it aloud and rewrite any phrase that sounds like a brochure.
• Is everything in it true? Check names, numbers and promises, because AI can state wrong things confidently.

Automate the repetitive steps, not the relationship

Automation goes a step further than drafting. A connected tool carries out a task each time something happens, without anyone starting it. Good candidates are steps that repeat, follow clear rules and can be checked by a person before anything reaches a customer:

• Turning meeting notes into an action list with owners and dates
• Sorting incoming enquiries by topic and drafting a first reply for someone to review
• Pulling figures from a spreadsheet into a weekly report draft
• Reminding the team when a follow-up is overdue

Keep a person at every point where a message leaves the organisation or a decision is made, including pricing, refunds, commitments and escalations. AI drafts. People decide.

Keep customer and company data in approved tools

Before pasting anything into an AI tool, check your organisation’s data policy. Personal data such as customer names and contact details falls under Malaysia’s Personal Data Protection Act 2010. Agree as a team which tools are approved and what must never be pasted into them.

What managers should do once the team starts using AI

Agree three rules with the team: which messages AI may draft, which it may not, and who approves anything automated that reaches a customer. Then review a few AI-assisted messages with each person every week or two. Look for generic phrases, missing specifics and facts nobody checked. The habit forms in these reviews, not in the first week of excitement.

What improves when your team uses AI this way

Routine writing takes less time, which leaves more time for work that needs judgement. Messages still sound like the person who sent them, and sensitive messages are written by people. Every automated step has a named person who approves what goes out, and customer data stays in approved tools.

How ClimbX supports this

ClimbX’s applied AI training follows the same split between AI drafting and human decisions. Participants build AI workflows they can use the next day, using real Malaysian work situations rather than imported scripts. They practise editing drafts and deciding which messages AI should not touch through role-play and group work, not lectures. After the workshop, reinforcement and manager coaching help the team keep the habit on real work. All ClimbX programmes are HRD Corp (HRDC/HRDF) claimable.

What to do next

If you would like to see where your team’s AI-assisted writing could be sharper, send one prompt or AI-drafted email your team currently uses to hello@climbxacademy.com. Remove any client names or pricing you prefer to keep private. We will return it with annotated feedback and show you where training would close the gap, so you can judge the value before deciding anything.