You can always tell. The AI-generated email that starts with "I hope this message finds you well" and ends with "Please don't hesitate to reach out." The one that's technically fine but has no personality. No warmth. No sense that a specific human wrote it to a specific other human.
That's the wrong way to use AI for client communication.
The right way keeps your voice and lets AI handle the parts that don't need it.
The Problem With "Write Me an Email"
When most people first try AI for communication, they open ChatGPT, type "write an email to my client about the project delay," and get back something that reads like a corporate form letter. So they either use it anyway (bad) or decide AI doesn't work for communication (also bad, and untrue).
The issue isn't the tool. It's the prompt. "Write me an email" gives AI nothing to work with except its training data, which is full of bland professional communication. Of course the output sounds generic. You gave it a generic instruction.
Prompts That Actually Work
Here's the difference. Instead of "write an email to my client about the project delay," try this:
"Draft an email to Sarah at Meridian Design. The website project is running two days behind because the content they were supposed to send last Friday hasn't arrived yet. Tone should be direct but warm. Don't blame them. Just note that we need the content by Wednesday to stay on schedule for the May launch. I usually write pretty casually, first names, short sentences, no corporate language."
That produces something you'd actually send.
The key ingredients are context (who, what, why), tone guidance (direct but warm, casual, no corporate speak), and constraints (don't blame them). The more specific you are about how you communicate, the better the output matches your voice.
Some people I work with keep a short "voice guide" saved as a note. A few sentences describing how they write. "I'm informal but professional. I use contractions. I don't use jargon. I keep sentences short. I sometimes use dry humor." They paste that into every communication prompt and the consistency goes way up.
The Three-Layer Approach
I recommend thinking about AI-assisted communication in three layers, depending on how much the relationship matters.
Layer one: fully automated. These are messages where personalization barely matters. Payment confirmations, appointment reminders, standard acknowledgments. "Got your form, we'll be in touch within 48 hours." AI writes and sends these. No human review needed.
Layer two: AI drafts, you edit. This is the sweet spot for most client communication. Proposals, project updates, follow-ups, responses to questions. AI creates the first draft based on your detailed prompt. You spend two minutes adjusting tone, adding a personal touch, and hitting send. You're editing, not writing from scratch.
Layer three: you write it. Sensitive conversations. Difficult feedback. Negotiations. Anything where the stakes are high enough that the message needs to feel genuinely human. AI can help you think through what to say (use it as a sounding board), but the writing should be yours.
Most businesses should have roughly 20% of their communication in layer one, 60% in layer two, and 20% in layer three. If you're trying to put everything in layer one, your clients will notice. If everything's in layer three, you're not saving any time.
Proposals and Scopes of Work
This is where AI saves the most time for service businesses, and where it's most tempting to cut corners.
A good proposal is part boilerplate and part bespoke. The boilerplate sections, your process overview, terms, pricing structure, company background, those don't change much. AI can assemble them. The bespoke sections, the problem statement, the specific approach, the "why us for this particular project" argument, need to feel written for this client.
The approach I recommend: build a proposal template in your AI tool of choice with your standard sections pre-loaded. For each new proposal, feed it your discovery notes (the actual notes you took during the call, not a sanitized version) and let it draft the custom sections. Then rewrite anything that sounds generic.
The discovery notes are the secret ingredient. They contain the specific language the client used, the problems they described, the things they emphasized. AI picks up on those details and reflects them back, which makes the proposal feel responsive rather than templated.
Follow-Ups That Don't Feel Automated
The biggest risk with AI-powered follow-ups is that they feel like AI-powered follow-ups. "Just circling back on my previous email" is the written equivalent of a robocall.
Better follow-ups reference something specific. "I was thinking about what you mentioned regarding the warehouse process, and I had another idea" is harder to ignore than "per my previous message."
To get this from AI, you need to feed it the context of the original conversation. Don't just say "write a follow-up to my proposal." Say "write a follow-up to the proposal I sent Maria last Tuesday. In our call, she was most interested in the reporting feature and mentioned they're under pressure to have something in place before Q3. Keep it short."
The result sounds like you actually remembered the conversation, because you did. AI just turned that memory into a polished paragraph faster than you would have.
A Note on Tone Across Cultures
If you work with clients across different cultural contexts, be careful with AI's default communication style. Most AI tools default to American professional English, which can feel too casual for some cultures and too formal for others.
Specify the cultural context in your prompt when it matters. "This client is in Germany and communication should be more formal than my usual style" produces meaningfully different output than "this is a casual Bay Area startup."
The Voice Test
Here's a quick way to check if your AI-assisted communication is working: read the email out loud. Does it sound like you? Would your client recognize your voice in it? If yes, you're using AI well. If it sounds like it could have come from anyone, you're not prompting it with enough of yourself.
The goal isn't to hide that you use AI. The goal is that it doesn't matter whether you do, because the communication is still genuinely yours.
For the admin tasks side of AI communication, like automated acknowledgments and reminders, that post covers the specifics. And if you want help setting up AI-assisted communication that actually matches your voice, we should talk.