I send a lot of emails. Client updates, internal proposals, difficult conversations, quick follow-ups — the range is wide and the volume is real. A few months ago I started using AI seriously to help, and it has made a genuine difference. But not in the way most people expect.
The biggest trap with AI-assisted email writing is using it to generate the whole thing from scratch and then sending it. That is how you end up with emails that technically say the right things but feel strangely flat — like something a competent stranger wrote on your behalf. Recipients notice, even if they cannot articulate why.
The method that actually works is different. Here is what it looks like in practice.
Step 1: You Write the First Draft (Even a Rough One)
Before I open any AI tool, I write a rough version of the email myself. Not a polished draft — just the core of what I want to say, in whatever form it comes out. It might be disorganized. It might be too long or too blunt. That is fine.
The reason this matters: the AI is going to improve what you give it. If you start with your own thinking and your own voice, the improved version still sounds like you. If you start with a blank AI-generated draft, you get something that sounds like a general-purpose professional communicator — which is no one in particular.
Step 2: Tell the AI What You Are Trying to Achieve
I paste my rough draft into Claude and give it context: who I am writing to, what the relationship is, what I need the email to accomplish, and what tone I want. Then I ask it to improve the draft — not rewrite it entirely, but improve it.
The context part is important. “Make this more professional” gives you something generic. “This is a follow-up to a client who seemed frustrated in our last call — I want to acknowledge the delay, explain what happened without making excuses, and give them a clear sense of what happens next” gives you something specific and useful.
Step 3: Read It Out Loud Before You Send
This is not an AI step, but it is essential. Before I send any AI-assisted email, I read it out loud at a normal pace. If any sentence makes me hesitate — if it sounds like something I would never actually say — I rewrite that sentence in my own words.
This catches the AI fingerprints that survive the editing process. Phrases that are technically correct but slightly off. Constructions that are smooth in text but awkward when spoken. A single pass of reading out loud catches almost all of them.
Specific Cases Where AI Email Help Is Most Valuable
Difficult conversations. When I need to push back on something, decline a request, or deliver bad news, having AI help me find the right framing saves a lot of agonizing over word choice. I write what I want to say, tell Claude the tone I am going for, and let it help me find language that is honest without being unnecessarily harsh.
High-stakes external emails. Anything going to a senior client, a potential new business contact, or someone I need to impress — I run it through Claude before sending. Not because I cannot write well, but because a second pass by something that has read a lot of professional communication tends to catch things I miss when I am close to the content.
Emails I have been putting off. Sometimes an email sits in my drafts for days because I do not know quite how to say what I need to say. Writing a rough version and handing it to AI to improve is often enough to break the blockage. The improved version usually shows me what I was trying to say, and from there editing is easy.
The One Rule
Your name is on every email you send. AI helps you communicate better — it does not communicate for you. Stay in the driver seat. Use it to sharpen what you are already trying to say, not to outsource the thinking about what to say in the first place.
Follow that rule and AI-assisted email writing makes you better and faster without making you sound like a machine. Ignore it and your recipients will notice something is off, even if they cannot name it.