AI
How to Use an AI Sales Email Writer Without Sounding Like a Robot
A practical way to write sales emails with AI that still sound like you, from the context you feed it to the claims you check before sending.
It's late. You just finished a good call with a prospect who wants to think it over, and you promised a recap email tonight. You open ChatGPT, type "write a follow up sales email," and get back something that opens with "I hope this email finds you well" and closes with "Don't hesitate to reach out." You would never say either of those things. Neither would anyone your prospect trusts.
That gap is what this article fixes. An AI sales email writer can save you real time, but only if you treat it like a fast junior assistant, not a ghostwriter. Here's how to set it up so the drafts sound like you, say only true things, and get quicker to produce every week.
Where AI helps and where it hurts
Start with an honest split. AI is good at some parts of email and bad at others, and knowing which is which saves you from cleaning up messes later.
Where it helps:
- Beating the blank screen. A rough first draft in ten seconds is worth a lot when you're tired.
- Organizing a brain dump. Paste your messy call notes and ask for a clear three paragraph email.
- Variations. Three subject lines, a shorter version for a text, a warmer version for a past customer.
- Tightening. Paste your own draft and ask it to cut 30 percent without losing the point.
Where it hurts:
- Fake warmth. Default output reaches for filler phrases that signal "template" from the first line.
- Made-up facts. Language models predict plausible text. Ask for an email about product results and it may invent a number that sounds right and isn't.
- Sameness. If everyone on your team uses the same prompt, prospects get the same email from five people.
- Over-polish. A sales email between two people should read like a note, not a brochure.
A simple rule: AI is good at form and bad at facts and feelings. You supply the facts and the feelings. It handles the shape.
Give the AI real context first
Most robotic emails come from thin prompts. When the model doesn't know something, it fills the gap with the average of everything it has seen, and the average sales email is bland.
Before you ask for a draft, give it five things:
- Who the person is. "Runs a home cleaning business with two employees. Found me through a webinar replay."
- What actually happened. "We talked for 20 minutes. She liked the scheduling piece. She's worried about price and about learning a new tool."
- The one next step. Not three options. One. "Book a 15 minute walkthrough this week."
- How you talk. Paste two or three emails you actually sent that got replies.
- What's off limits. "No income claims. No fake urgency. No 'limited time.'"
The voice samples matter most. OpenAI's own prompt engineering guide recommends clear instructions and examples of the output you want. Your old emails are those examples. Pick ones where you sounded relaxed and got a response, not the ones you agonized over.
Here's how that plays out for a network marketer. Say you mailed a product sample to someone you met at a local market, and they texted "got it, thanks." Thin prompt: "Write a follow up about a sample." Rich prompt: "She runs early morning bootcamp classes, mentioned her clients always ask about recovery, got the sample Monday, said thanks by text. I want to ask if she's tried it yet, nothing pushy." The second prompt gives the model something real to work with, and the draft will show it.
A prompt structure that works
You don't need clever tricks. You need the same structure every time, so you can fill it in fast from your phone. Here's one you can copy into a notes app:
Role: You are helping me write a short follow-up email to a prospect. Write in my voice, using the samples below.
About the prospect: [2 to 3 sentences]
What happened: [what we discussed, with their concerns in their own words]
Goal of this email: [one next step]
My voice samples:
[paste 2 to 3 real emails]
Rules:
Keep it under 120 words.
Do not open with "I hope this finds you well" or similar.
No exclamation points unless my samples use them.
Do not state any numbers, results or claims that are not in this prompt.
End with one simple question.
Give me 2 versions: one direct, one warmer.
Each line earns its place:
- The role line tells the model it's helping you, not writing as a generic salesperson.
- Their concerns in their own words keeps the email about the prospect. If she said "I'm not techy," put that in quotes. The model will answer the real worry instead of a guessed one.
- One goal stops the email from turning into a menu. One ask gets more replies than three.
- Under 120 words because most people read email on their phone. Long blocks get skimmed or saved for later, and later rarely comes.
- The no-new-claims rule is your first guardrail against invented facts. It won't catch everything, but it helps.
- Ending with a question gives the reader an easy way to reply.
- Two versions let you pick the closer fit instead of rewriting from scratch.
Then push back. "Use version two but shorter." "The second sentence sounds stiff. Make it how I'd say it out loud." Two or three quick rounds usually get you close.
Edit for your own voice
The draft is not the email. Read it out loud once. Anything you'd stumble over saying, change. Then run this quick pass:
- Cut the first sentence. It's often throat clearing. The real opener is usually sentence two.
- Swap fancy words for plain ones. "Utilize" becomes "use." "Reach out" becomes "text" or "call."
- Add one detail only you know. Her dog barking halfway through the call. The soccer game he was rushing to. That detail is the one thing no AI can generate, and it's what makes the email feel personal.
- Check it on your phone. If it scrolls more than once, cut.
- Fix the sign-off. Use whatever you normally write, even if it's just your first name.
This should take about two minutes. If editing takes longer than writing from scratch would, your context was too thin. Go back and add more to the prompt next time.
Check every claim before sending
This step keeps you out of trouble. Every number, every product benefit and every "most people find" needs to be something you can back up.
The FTC's business guidance on multi-level marketing is clear that earnings claims and product claims must be truthful and not misleading. It doesn't matter that a machine wrote the sentence. You hit send, so you own it.
If you take an approved draft and send it to a list, other rules come into play. The FTC's CAN-SPAM compliance guide covers honest subject lines, identifying yourself, including a physical address and honoring opt outs. Which rules apply depends on the primary purpose of the message, so read the guide before you load anything into a bulk sequence. Inbox providers have their own requirements too. Google's email sender guidelines explain authentication and unsubscribe expectations for people sending to Gmail addresses.
Before sending, scan for:
- Any number. Where did it come from? If you can't point to a source, delete it.
- Any product benefit. Does it match your company's approved language?
- Any hint about money or lifestyle. "Imagine the freedom" counts. Delete it.
- Any promise about timing. "I'll send it tomorrow" only works if you will.
- Personal details. If you pasted notes on several prospects, the model may mix them up. Check names, products and what was said.
Build a small library of approved drafts
After a few weeks you'll notice you keep writing the same handful of emails. For most sellers it looks something like this:
- After a first call
- After a no-show
- After "I need to think about it"
- After a sample or trial goes out
- After two weeks of silence
- A check-in with an existing customer
For each situation, save one edited, claim-checked draft plus the prompt that produced it. Label each with when to use it. A notes app works. A shared doc works. Templates inside your CRM work best because they sit next to the contact record.
Now your process changes. Instead of starting cold, you paste the approved draft into the AI with today's details and say "adapt this for this person." The structure and the claims are already safe. You're only changing the personal parts. That's faster and safer than starting over.
Team leaders get an extra benefit. Review the library once, make sure the claims are clean, and share it. Then have each person swap in their own voice samples so your team doesn't all sound identical. Once a month, look at which drafts get replies and retire the ones that don't. If you want to pair the library with a timing plan, a written follow-up schedule tells you which draft to pull and when.
Common questions
Will prospects be able to tell an email was written by AI?
They can usually tell when an email is generic, and generic is what AI produces with thin context. Feed it your real voice samples and the details of the conversation, then add one personal detail by hand. That combination reads like a note from a person because it mostly is one.
Is it okay to use ChatGPT for sales emails in a direct selling business?
Generally yes, but check your company's policies first. Many direct selling companies have rules about approved product language and claims. Whatever tool drafts the email, you are the sender, so you are responsible for every claim in it.
Should I let AI send follow-up emails automatically?
For routine messages like booking confirmations or reminders, automation can work well. For anything that responds to a prospect's specific concerns, review the draft before it goes out. Those are the emails where a wrong detail or a made-up claim costs you trust.
The bottom line
AI won't make you a better salesperson, but it can give you back the twenty minutes you spend staring at a blank email every night. The trick is the order of operations: real context in, a consistent prompt, a quick voice edit, a hard claim check, and a growing library of drafts you already trust. Do that for a few weeks and your follow-ups get faster without getting colder.
If you'd like your templates, contact notes and follow-ups in one place instead of spread across five apps, Prala keeps them together with AI drafting built in. You can see the plans here or start an account when you're ready.
