I Asked ChatGPT to Help Me Write Client Emails — Here's the Exact Prompt That Actually Worked

I spent 25 minutes writing one client follow-up email last spring. One email. To one person. I knew what I wanted to say, I just couldn't get it to come out right without sounding either too stiff or too apologetic or weirdly aggressive. So I rewrote it four times, hated all four versions, sent the third one anyway, and then thought about it for the rest of the afternoon.

That is not a workflow problem. That is a tax on being a person who runs a service business and actually cares how she comes across. And if you've ever stared at a blank reply to a client email for longer than it took to do the actual work, you already know exactly what I'm talking about.

ChatGPT email prompts for service businesses get talked about constantly in AI circles. The advice you find is almost always the same: "Ask ChatGPT to write your emails for you!" Then they hand you a generic prompt that spits out something that sounds like a corporate HR department drafted it in 2019. You paste it in, wince a little, and send it because you're already late. The client notices something is off, even if they can't say what.

This article is about what actually worked, after the generic prompts didn't. The exact prompt. The reasoning behind it. And why the fix was smaller and weirder than I expected.

The Real Problem Isn't Writing. It's the Setup.

Here's what nobody says when they hand you a ChatGPT prompt for emails: the output is only as good as the context you give it. If you type "write a professional follow-up email to a client who hasn't paid," you will get a professional follow-up email to a client who hasn't paid. It will be correct. It will also sound like nobody you've ever met, and your longtime client will read it and feel like they accidentally emailed the wrong business.

The problem isn't the tool. The problem is that you handed it a task instead of a situation. ChatGPT doesn't know your client's name, doesn't know you've been working together for three years, doesn't know that she always pays eventually but sometimes goes quiet for a few weeks because she runs a chaotic household, and doesn't know that you'd rather sound warm than firm even when you're a little annoyed.

That missing information is everything. Strip it out and you get a technically correct email with no soul in it. And in a service business, the email is part of the service. Clients feel the difference even when they can't name it.

This is the same reason most people give up on AI writing tools after one or two tries. The first draft comes back wrong, they assume the tool is bad, and they go back to writing it themselves for 25 minutes. What they actually needed was to change what they fed it, not switch tools.

What I Tried Before It Clicked

I'll be honest about the bad prompts I used before I landed on what worked, because they're probably the same ones you've tried.

Version one was basically just a command. "Write a follow-up email to a client about their overdue invoice." The output was fine. It was also the exact email you'd get if you Googled "overdue invoice email template" and clicked the first result. Nothing wrong with it. Nothing right about it either. It didn't sound like me, it didn't reflect anything about the actual situation, and I rewrote most of it anyway.

Version two, I tried being more specific about tone. "Write a follow-up email that's firm but friendly." ChatGPT interpreted "firm but friendly" as a specific personality type that apparently went to business school and enjoys team-building retreats. Still not it.

Version three, I found someone online recommending that you paste in a sample of your own writing first. That was closer. The tone got better. But it still wasn't accounting for the relationship or the specific situation, so the email felt like me in the wrong context, like I was sending a message that fit a different client to this one.

None of those failed because ChatGPT is bad at email. They failed because I was still treating it like a vending machine. Put in a request, get a result. That's not how it works when the output actually needs to fit a real human on the other end.

The Reframe: Stop Asking It to Write. Ask It to Think First.

Here's the shift that changed everything for me, and it maps directly to how I built Richard's website and how I rewrote Jo's copy. I stopped starting with the output and started starting with the person.

With Richard's site, I didn't open a page builder and start typing services. I thought about who actually calls a guy who fixes trucks out of a garage, and I built backward from that person. Tradesmen don't check email, so I hooked up text. I wrote the location copy the way someone would say it out loud, not the way you'd stuff a keyword. The site worked because it was built for the actual human, not for the algorithm.

The same logic applies to every email you're staring at. You're not writing an email. You're writing to a specific person, in a specific situation, on behalf of a specific version of yourself. ChatGPT can do that. But you have to actually tell it all three of those things before you ask it to write a single word.

When I made that switch, the quality of the output changed immediately. Not because the prompt was longer. Because it was specific in the right direction.

The Exact Prompt That Worked (And Why Each Part Is There)

Here's the prompt framework I use now for client emails. I'm going to give you the structure and then explain what each piece is actually doing, because copying a prompt without understanding it means you'll be stuck again the next time the situation is slightly different.

I need to write an email to [client name or description]. Here's the context: [2-3 sentences about who they are, how long you've worked together, anything relevant about the relationship]. The situation is: [what happened, what you need them to do, any relevant details]. My tone is usually [describe in plain words — warm but direct, conversational, professional but not stiff, etc.]. I don't want it to sound like a template. Write a short email, under 150 words, that sounds like a real person wrote it.

That's it. Let's talk about why it works.

"Here's the context: [who they are, how long you've worked together]" — This is the part everyone skips. It's also the part that makes the output actually usable. ChatGPT can calibrate warmth, familiarity, and stakes when it knows the relationship. A three-year client and a brand-new inquiry do not get the same email, even if the topic is identical. Telling it the relationship changes the whole draft.

"The situation is" — Not "write an email about X." That's vague. This is specific: what actually happened, what you need, what the context is. The more honest and real this part is, the better the draft. If you're a little annoyed, say so. If there's a reason you've been delayed in following up, include it. Give it the real situation, not the sanitized version.

"My tone is usually [plain description]" — Skip words like "professional" and "friendly" because they mean nothing to ChatGPT. Instead, use comparisons or descriptions: "like texting a client I know well," "direct but not cold," "warm enough that they don't feel like a transaction." The more specific the description, the closer the draft lands.

"Under 150 words" — This is a hard constraint, and it is your friend. Without it, ChatGPT writes four paragraphs when you needed two sentences. Short emails get read. Long emails get skimmed and sometimes just create more confusion. Force the word limit.

"Sounds like a real person wrote it" — This one sounds obvious and makes a real difference. It tells the model to skip the corporate-formality defaults and write like a human is actually talking to another human. Weirdly effective.

How to Use This in Under 15 Minutes

This is where the 15-minute workflow actually lives. The goal isn't to spend an hour crafting the perfect prompt. The goal is to spend two minutes filling in the framework, get a usable draft back in 30 seconds, and spend another few minutes editing rather than writing from scratch.

Here's what a full session looks like in practice.

Open a fresh ChatGPT conversation. Paste the framework above, with your details filled in. Read the first draft. It will probably be 80% right. The 20% that's off is usually fixable with one follow-up message: "Make it a little shorter," or "Drop the last sentence, it sounds like a form letter," or "The opening is too formal, make it sound like I'm just checking in." Send your follow-up. Read the second draft. That one's usually good enough to send. Total time: under 10 minutes, including your edits.

The five minutes you save doing other things is bonus. The real gain is not carrying that email around in your head for the rest of the afternoon.

For repeat email types, the workflow gets faster. Once you've written a good "first reminder on a late invoice" prompt for your business, save it somewhere. Next time you need it, swap out the client name and the specifics, paste it in, and you're done in three minutes. That's the version of AI that's actually useful: the one that removes a recurring friction point, not the one you have to set up for an hour before it saves you twenty minutes.

If you want to go deeper on building these kinds of repeatable workflows, this breakdown on building simple AI workflows for service businesses walks through the same logic applied to your whole operation, not just email.

Does This Actually Sound Like You?

Here's the one thing I'd push back on if someone told me AI-written emails are fine to send without editing: don't skip the editing step.

Not because ChatGPT gets it wrong, but because a client who's known you for two years knows how you write. They notice when your sign-off changes. They notice when your email is suddenly longer and more formal than usual. The draft is a starting point, not a finished product. Your job in those last few minutes is to read it out loud in your own voice and fix anything that doesn't sound like you said it.

One line you'd never say in real life? Cut it. A phrase that's technically correct but sounds like a template? Swap it for how you'd actually say it. Sometimes that means adding a short line that's just you being human: "Hope your week got easier" or "Let me know if you have any questions, I'm around." Those lines don't improve the information in the email. They preserve the relationship.

That is something a prompt can't do for you. But once the heavy lifting is done, it only takes a minute.

This connects to something I wrote about in talking to ChatGPT like a business owner: the whole point is to show up in the conversation like a person with a specific situation, not a person with a generic task. The prompt is just that principle applied to email.

What Kinds of Emails This Works For

I've used this framework for every recurring email situation that used to slow me down. Late invoice follow-ups. Rescheduling requests. Checking in after a big service to see how a client is feeling. Saying no to something without sounding rude about it. Responding to a complaint without being defensive. Welcoming a new client. Following up on a quote that went quiet.

The framework holds across all of them because the logic is the same. Context first, situation second, tone third, constraint on length. What changes is what you fill in. That's why it's worth learning the structure instead of just copying one specific prompt. One prompt solves one problem. The framework solves every version of this problem, including the ones you haven't hit yet.

If your business has a heavier client communication load, like ongoing client check-ins or a formal onboarding process, the next logical step is a full follow-up workflow. This piece on building a client follow-up workflow goes deeper into the systems side of it.

The Honest Reality Check

ChatGPT email prompts for service businesses are useful. They are not magic. They don't replace knowing your client. They don't replace the two minutes you spend skimming the draft before you hit send. And they don't help at all if you're still treating AI like a vending machine where you drop in a vague request and expect a perfect product to come out.

What they do replace is the 25-minute stare at a blank reply box. The rewriting something four times because you can't get the tone right. The mental weight of carrying "I still need to send that email" around all afternoon. That part, the friction part, AI is genuinely good at removing.

The rest is still yours. Which, honestly, is fine. The relationship is yours. The client is yours. The email is just the logistics.

Start This Week

Pick one email you've been putting off. Not the hardest one, just one you've been avoiding. Open ChatGPT, paste in the framework with your details, and send what it gives you after one round of editing. That's the whole assignment.

If you want more workflows like this one, every week on Substack I break down a new 15-minute AI workflow for real service business situations, written in plain English, no jargon, no hype. Subscribe at Simply Doing AI and the next one lands in your inbox before you have to go looking for it.

Frequently Asked Questions

What are the best ChatGPT email prompts for a service business?

The most effective ChatGPT email prompts for service businesses include the relationship context, the specific situation, a plain-English tone description, and a hard word limit. Generic prompts that just name the email type produce generic results. The more specific your input, the more usable the draft.

Will my clients know my emails are AI-assisted?

Not if you edit before you send. The draft is a starting point, not the finished email. Read it out loud in your own voice, swap any phrases that don't sound like you, and add one human line if needed. That edit step is what makes it yours.

How long does it take to get a usable email draft from ChatGPT?

Under three minutes once you have the prompt framework ready. Fill in the context, paste it in, and the first draft comes back in seconds. One or two follow-up messages to refine it and you're done. The whole process, including your edits, should take less than 10 minutes.

Can I use ChatGPT email prompts for difficult emails, like late payment reminders or complaints?

Yes, and those are actually where the framework helps most. Difficult emails are the ones we rewrite six times and still feel unsure about. When you give ChatGPT the full situation, including the awkward parts, it produces drafts that are firm or sensitive in the right proportions. You still edit, but you're starting from something usable instead of a blank page.

Do I need a paid ChatGPT subscription for this?

The free version of ChatGPT handles email drafts well. Paid versions (ChatGPT Plus) give you access to more advanced models that tend to nail tone more consistently, but the core framework works on either. Start with whatever you have access to.

What if the first draft still sounds off?

Send one follow-up message that names exactly what's wrong: too formal, too long, last paragraph doesn't sound like me. ChatGPT adjusts well when you give it specific feedback instead of just asking it to "make it better." Two rounds of revision is usually all it takes.