I Let AI Write My Client Emails for 30 Days — Here's What Happened
I want to be clear about what I mean by "let AI write my emails."
I did not give an AI access to my inbox. I did not automate the sending. I did not remove myself from the loop.
What I did: every email that required more than a one-line reply, I drafted using AI first, then reviewed, edited as needed, and sent. For thirty days.
This is what happened.
Why I Started
I was spending between 90 minutes and two hours every day on email.
Not because I receive an extraordinary volume. Because I was writing carefully — reading the email, thinking about the right response, drafting, reading back, adjusting. For every email.
I'm a consultant. My emails are often where the relationship lives. Getting them right matters.
But two hours a day on email when you're a solopreneur is time not spent on billable work, content, or the things that actually grow the business.
I wanted to know if AI could help without making my communication feel cheaper.
The Setup
I used primarily Claude for longer, more nuanced emails — difficult client situations, proposals, sensitive responses.
I used ChatGPT for shorter emails — quick updates, scheduling, confirmations, standard follow-ups.
My process: paste in the original email, describe the context and what I wanted to achieve, generate a draft, review, adjust, send.
I tracked: time spent per email, how much editing was required, and any situations where the AI draft caused problems.
Week One: The Learning Curve
The first week was slower than doing it manually.
I was still learning how to prompt well. My first attempts produced generic, slightly too-formal emails that didn't sound like me. I spent almost as long editing as I would have spent writing.
Key lesson from week one: AI writes in your voice when you tell it your voice.
By day five I had developed a short context block I pasted before every prompt: my name, my tone (direct but warm, professional but not corporate), my client's name, and what the relationship status was. Results improved significantly.
Week Two: The Efficiency Kicks In
By week two, my average email drafting time had dropped from 12 minutes per email to about 4 minutes.
Most drafts needed one or two sentence adjustments before sending. The AI was capturing my context well, and the prompts were getting better.
The emails that worked best: follow-up after a meeting, proposal summaries, update emails, responses to questions I'd answered before in slightly different forms.
The emails that needed significant editing: anything emotionally charged. A client who was frustrated about a delay. A conversation where the subtext was complicated. The AI would produce technically correct responses but miss the human layer.
Week Three: The Embarrassing Moment
I want to be honest about this.
On day 19, I sent an email to a long-standing client using an AI draft I hadn't reviewed carefully enough.
The email was professionally fine. But it started with "I hope this email finds you well" — a phrase I never use, have always found slightly hollow, and which my client had actually joked about hating in a conversation months earlier.
She replied: "I hope it does too! Did something happen to your email? This doesn't sound like you."
It was a light moment, not a disaster. But it was a reminder that AI doesn't know your relationship history. It doesn't remember that this particular client prefers blunt directness. It doesn't know that you always check in about her daughter's university applications.
The context you don't provide is invisible to the AI.
Week Four: The New Normal
By week four I'd built a workflow that worked.
For most standard emails: AI draft → 60-second review → send. For complex or relationship-sensitive emails: AI draft → careful read + edit → sometimes a complete rewrite using the AI's structure as a scaffold.
My average daily email time had dropped from 90-120 minutes to 45-60 minutes. That's real time back in my week.
The Honest Results
Time saved: approximately 40% reduction in email drafting time.
Quality impact: Better for some, same for most, occasionally worse when I didn't review carefully enough.
Client reactions: Two clients mentioned my emails seemed crisper recently. One noticed when I sounded uncharacteristically formal.
Mistakes made: Two emails sent without sufficient review. Neither caused damage. Both taught me something.
The one thing I didn't expect: AI first drafts often surface the clearest way to say something. My natural drafting can be meandering. AI tends to be more direct. Reading the draft sometimes showed me that the direct version was better than whatever I would have written.
What AI Can and Cannot Do for Client Email
AI can:
- Draft clear, professional responses quickly
- Match tone and formality to context when prompted
- Handle standard situations well
- Give you a starting point that's 80% there
AI cannot:
- Know your relationship history with the client
- Pick up on emotional subtext
- Know the things you never wrote down
- Replace your judgment about what the situation actually requires
The Verdict
Worth doing. With caveats.
The productivity gain is real and consistent. The risk of the emails feeling less like you is also real and requires active management.
My recommendation: use AI drafts for all standard, non-sensitive emails. Write sensitive, relationship-critical emails yourself — or use AI for the structure and write the parts that matter yourself.
The 40% time saving across email is not nothing. Used consistently, that's hours per week returned to you.
Try It This Week
Pick five emails from your inbox that need replies. Use Claude or ChatGPT to draft each one. Review, edit, send. Track your time.
You'll know within a week whether it works for you.
Watch how I set up my email prompting workflow on the Automate Lane YouTube channel: Automate Lane on YouTube
Automate Lane runs real experiments so you can make informed decisions about AI tools. Subscribe for weekly results.