You will be able to steer the voice of the output with samples of your own writing and examples of what not to do.
Daniel's assistant now tags messages well, so he has started asking it to draft the replies too. They are polite and accurate, and they sound nothing like him. Every reply opens with "Thank you so much for reaching out", uses words like "rest assured", and closes with "Is there anything else I can help you with?" His regular customers notice. One of them replied to a draft he forgot to edit: "Daniel, is this a bot now?"
Lesson 2.1, One good example beats a paragraph of description, showed that examples set the pattern for structure and content. Voice works the same way, with a couple of extra moves that help a lot.
Telling an assistant to write "in a friendly, down-to-earth way" gets you its idea of friendly and down-to-earth, which is a fairly generic one. Pasting two or three things you actually wrote gives it the real thing: how long your sentences run, which words you use, how formal you are, whether you use contractions, how you open and how you sign off.
Choose samples that match the job. If you want customer replies, give it customer replies. A LinkedIn post and a condolence message are both your writing, but neither tells the model how you answer a question about delivery dates. Pick pieces you were happy with and that were written without much editing, since those are closest to your natural voice.
Then be specific in the instruction. "Match the sentence length, vocabulary and level of formality of these samples" works better than "write like me", because it tells the model which features to copy and leaves the content to the new task. Remove names, order numbers and anything else confidential from the samples first. The model needs your voice, not your customers' details.
Here is a step most people skip. Before asking for a draft, ask the assistant to describe the style it sees in your samples, in a short list. Something like: "Before you write anything, describe the writing style of these three emails in five bullet points."
This does two things. It shows you what the model has picked up, and it gives you a chance to correct it before it writes a word. When Daniel tried it, the description said his style was "casual, with frequent humour". He does not think of himself as funny. Looking again, he saw that one of his three samples was a joking reply to a customer he has known for years, and the model had treated that as typical. He swapped that sample for a more ordinary one and corrected the description: "friendly but brief, no jokes with customers I do not know".
Once the description is right, save it. A few corrected lines describing your style are short enough to paste into any request, and later you can put them into custom instructions or a project so they apply automatically, which module 6 covers.
Sometimes the model keeps sliding back to a habit even after you have given it good samples. This is where a counter-example helps: a short piece of text, clearly labelled as what not to do.
Daniel added this to his prompt: "Here is an example of the tone I do not want. Avoid writing like this:" followed by one of the over-polished drafts the assistant had produced earlier. The label matters. Without it, the model may treat the counter-example as one more sample to copy, which is the opposite of what you want. With it, the model has a clear contrast between the samples it should match and the one it should move away from.
Keep the counter-example short and make sure it is genuinely the thing you dislike. One paragraph is enough.
The most reliable fix is often the plainest one: name the specific habit and ask for it to stop. Vague instructions such as "less formal" or "more natural" leave the model guessing what you mean, while a named habit is something it can check its own draft against.
Daniel's list ended up as four instructions:
Do not open by thanking the customer for reaching out. Do not end with a question unless I need an answer from them. No "rest assured", "kindly" or "we apologise for any inconvenience". Use the customer's first name once, at the start, and nowhere else.
Your list will be different. Common ones include opening with a compliment, repeating the question back before answering it, ending every email with a question, stacking three adjectives together, and adding a summary paragraph that says everything a second time. Read three or four drafts the assistant has written for you and note what made you wince. Those are your habits to name.
Put the pieces together and Daniel's request now has his corrected style description, two sample replies, one labelled counter-example and his list of habits to drop. His replies still need checking for accuracy, but they rarely need rewriting for tone.
Getting there starts with an honest look at your own writing. Go to your sent folder and pick three emails you wrote quickly to people you deal with regularly, the kind you would be comfortable having an assistant learn from.
Paste three of your own emails, ask the assistant to describe your style in five bullet points, correct anything it got wrong, and save the final description.
Junxiong-WFG Organisation is an authorised representative of AIA Financial Advisers Private Limited (Reg. No. 201715016G).