Writing Customer Emails and WhatsApp Messages with AI: Tone, Speed and Limits

Writing customer emails and WhatsApp messages with AI is the skill that pays off fastest for sales, customer service, real estate, tourism and office workers who produce dozens of messages a day. ChatGPT and Claude give you the draft within seconds, but the secret of a good message lies in three things: the right channel, the right tone and one sentence specific to the customer. Without these three the message reads like a robot and loses trust.

In this article I will explain the patterns we build most often in individual sessions, the channel difference and the data limit.

The channel difference: email and WhatsApp are not the same thing

Email is structured: greeting, context, information, next step, closing. WhatsApp is short: two sentences, one question. The pattern for those who want to use the same content on both channels: "Write this information in two versions: email (5 to 7 sentences, structured, formal but warm) and WhatsApp (2 sentences, friendly, ending with a single question). In both, leave one field to be filled in specifically for the customer."

Tone: teach the tool your own voice

Put three good messages you wrote yourself at the top of the prompt and say "write in this tone". This single step changes output quality more than anything else. Without a tone example the tool produces corporate advertising language. For a team the same job is done with a brand voice document. Details are in the article on an AI workflow for marketing teams.

Patterns for five frequent situations

• A customer asking for information: "Write a reply to this question: [question]. First the direct answer, then one piece of additional information, then the next step. Do not make promises, and use 'at the latest' when giving a timeframe."
• Appointments and confirmations: "Write a confirmation message for this appointment: [day, time, place or link]. If preparation is needed, add a single item. Leave a polite exit sentence for cancellation."
• Delay notifications: "Write a message announcing the delay: acknowledgement, reason (short), new date, and mention compensation if there is any. Do not make excuses, do not exceed three sentences."
• Follow-up sequences: "Write a three-step follow-up for a proposal that went unanswered: day 3, day 10 and day 21. Each message two sentences, without pressure, ending with a single question."
• Complaint replies: "Write a reply in the acknowledge, apologise, solve, timeframe and invite structure. Do not start an argument, give two tone options."

Personalisation: one sentence is enough

Every message should have a single sentence specific to the customer: a reference to the last conversation, the name of their project, the concrete question they asked. The tool leaves this field empty and you fill it in. This sentence is the only thing that proves the message is not a template, and it takes ten seconds.

The data limit

The customer's name, phone number, order and payment details do not go into the prompt. The context is written anonymously ("an e-commerce customer asking about a shipping delay") and the personalisation field is filled in by hand before sending. If you work in a company, a corporate account and a written data rule are essential. Details: AI data security and privacy training.

How to start

Start with a week of your own messages: pick out the 10 types of message you write most often, build a pattern for each, add your tone example and try it for a week. To build this together in 90 minutes, see individual AI training. For a team, the sales teams or customer service teams program. The basic patterns are in the free prompt library.