Last week I dumped every client message on AI. Customer #9 refunded.
I was racing a deadline, so I let AI reply to my client messages for me. The first 8 went fine. Customer #9 refunded outright, with the note: "Your replies are too 'correct' — feels like a mass-send." That's when it hit me. I'd been treating AI like a "coworker who decides things for me."
AI isn't deciding for you. It's just "talking back."
I recently came across an article called "Computers Cannot Make Decisions." One line in it gave me chills: computers (including today's AI) only execute instructions and predict the next most likely word. They never "decide." They hand you a paragraph, three options, a draft reply — but whether to use it, tweak it, or hit send, that call is yours.
Lin Hui (a.k.a. Lao Lin / Xiaohui), a content creator, has been leaning on AI to write his WeChat Moments copy for the past year. Now he rewrites every AI draft — mood, details — before posting. His line: "AI gives me the bones. The flesh is mine to grow." Afei, an indie developer, takes a different route for client emails: he feeds AI his client's last 3 chat messages, asks for a draft, then reads every word himself to check the tone before sending.
What I'm doing today: take back the send button.
Money: $0 (you're already on ChatGPT, Wenxin, Kimi or similar, right?)
Time: 30 minutes today
Tech barrier: no code — just pause 5 seconds before hitting "send" and ask yourself, "Am I speaking, or is the AI?"
First step: Open the AI tool you use most. Type in your next message. Before you hit "copy" or "send," pause 5 seconds. Read it. Ask yourself, "Does this line need a tweak?" The decision is yours.
Wherever you are, here's how I'd try it.
Just starting out (0 clients / still making content): When I was here, I didn't let AI talk to my clients for me. I used it to outline, brainstorm angles, look things up — but I kept the writing and the tone calls for me. Honestly, training that "does this AI output feel right?" instinct has paid off more than any productivity trick I've tried.
Got 1–2 clients now: This is when I started adding one line only I'd say — a specific detail, an inside joke, a "let me think on this." AI can't learn that part. And clients notice when it's missing.
Scaling up (3+ people / bigger projects): This is where I'm at now — splitting the work: AI drafts, a human reviews. AI takes speed. We keep the tone, the judgment, the accountability. The last thing a client reads should always have a person's name behind it.