The number one fear executives have about AI-written email is simple: "it will not sound like me." They are usually right, because most AI email sounds like everyone. But that is a fixable problem, not a reason to avoid the tool.
Why AI email sounds generic
AI cannot guess a voice. It can only match one you show it. Ask for a reply with no examples and no rules, and you get the flat, over-polished default. The fix is to give it real samples and clear guardrails.
The prompt I use
"You are drafting on my behalf. Here are three emails I have written: [paste]. Notice the tone, length, and sign-off. Draft a reply to this, matching that voice exactly. Keep it under [X] words. Do not invent facts; leave a [bracket] wherever you need input: [paste]."
The two lines that matter most
"Do not invent facts" and "leave brackets for anything unsure" are what turn a risky tool into a safe one. They stop the model from confidently filling gaps with things that are not true, which is exactly where AI email gets people in trouble.
Always do the final read
Even a great draft gets a human read before it sends, especially anything sensitive. The prompt saves you time; your eyes save your executive's credibility. That is the whole trade, and it is a good one.