Having artificial intelligence write email drafts for your team is easy. Getting it to write emails that sound like your firm, and not like every other firm using the same model, is the hard part, and it is the question this article answers. The answer is less about the tool and more about what you give it: a brief, real samples of your writing and a review rule. Without those three, AI produces polite, generic email that your prospects recognise within a sentence and delete.
Why AI email sounds generic
A language model writes the most likely next sentence given what it has been told. If it has been told only "write a follow-up email to a prospect", the most likely email is the average of millions of follow-up emails. That average opens with "I hope this finds you well", restates the obvious and closes with "let me know if you have any questions". It is not wrong. It is just indistinguishable.
Your voice is the set of choices that make your email less average: how short you write, what you lead with, which words you never use, how you sign off. The model cannot guess those choices. You have to give them to it.
The three inputs that change the output
A brief for each email type. One paragraph stating who receives this email, what they already know, the one thing it must achieve and what the reader should do next. A brief for a first reply to an enquiry is different from a brief for a proposal chaser.
Voice samples. Three to five real emails your best writer sent, which got replies. Not your marketing copy, which is usually more formal than your people. Real one-to-one emails. Paste them into the AI instructions as examples of the voice.
Rules. A short list of what never appears: phrases you dislike, claims you cannot make, the maximum length. Rules are where you stop the model drifting back toward the average.
Here is what such a rule set might look like. Maximum 120 words for any one-to-one email. Open with the reason you are writing, never with a pleasantry. One question per email. No "just checking in", "circle back" or "touch base". Sign off with first name only. Never mention pricing in a first reply. The agency would add these six lines to every AI writing prompt.
One-to-one replies versus campaign emails
These are different jobs and they need different set-ups.
One-to-one replies are answers to a specific person: an enquiry, a question about a proposal, a scheduling request. AI drafts, a person reads and sends. The draft should use what the contact record says, such as their company, what they asked and who they spoke to. Speed matters more than polish, and the review takes seconds. This is where AI saves the most time for a small sales team. For the follow-up sequence that comes after, see AI follow up.
Campaign emails go to a list: a newsletter, a reactivation campaign, an event invitation. Here AI helps with drafts, subject line options and variations for different segments. Each email is read by hundreds of people, so a person edits more heavily. Mistakes are multiplied, so the review is slower and deliberate.
The trade-off is between speed and control. One-to-one drafting gives you speed with light review because the stakes per email are low. Campaign drafting gives you a faster first draft but keeps the editing time, because one weak sentence reaches the whole list.
Before you start
- Three to five real emails that represent how your firm writes at its best.
- A one-paragraph brief for each email type you send often, usually four to six types.
- Your voice rules, written as short lines.
- Contacts in the CRM with source, company and conversation history, so drafts can be personal.
How to set up AI drafting
- Go to Marketing > Emails > Email Campaigns and click Create Campaign.
- In the Email Builder, drag in or edit a Text Block.
- Click the Content AI button above the text editor.
- Choose a context from the Select Context dropdown.
- Enter a Content Title and a short Content Description.
- Select a Writing tone and the Number of Variations.
- Click Generate.
- Click Copy next to the variation you want, then Continue.
The exact wording of the AI options varies by screen, but the principle is the same everywhere: give the instructions once, at the level where they apply to every draft, rather than retyping them for each email.
Where it goes wrong
Invented facts. AI fills gaps. Ask it to follow up on a proposal without telling it what the proposal contained, and it will describe one. Always include the facts in the brief or make sure the contact record holds them. The review rule should check every specific claim.
Wrong tone for the relationship. A draft for a long-standing client should not sound like a draft for a cold prospect. Use different briefs for different relationships, and let the draft read the contact's history.
Length creep. Models like to explain. Without a word limit, a three-line reply becomes a seven-line one. Put the limit in the rules and enforce it in review.
Sameness across the team. If five salespeople use the same prompt, prospects who speak to two of them will notice identical phrasing. Encourage each person to add a personal sample to their own drafts, or at least to change the opening line.
Suppose advisers take a median of four hours to reply to new email enquiries, because each reply is written from scratch between client meetings. With AI drafts using the firm's brief and rules, an adviser could review and send a reply in under two minutes. If that brought the median reply time to around 35 minutes, with advisers editing about a third of drafts to add a detail from memory, the gain would come almost entirely from removing the blank page.
A review rule that holds
Agree one rule for every AI-drafted email before it is sent: the sender has read every sentence and could defend it to the recipient. That sounds obvious. In practice it is the step people skip once drafts start looking good, and it is the step that catches the invented proposal detail or the promise you cannot keep.
For campaigns, add a second reader. For one-to-one replies, the sender is enough. Do not allow fully automatic sending of AI-written email to prospects or clients until you have reviewed several hundred drafts of that type and found them consistently right. Even then, keep automatic sending to low-stakes messages such as booking confirmations.
What most guides on AI email writing get wrong
Most guides focus on prompt tricks: act as a sales expert, use a friendly tone, add urgency. Those instructions push the model toward the average, because they describe the average salesperson. What moves output toward your voice is your own material: real samples, real rules, real facts from the contact record. A dull prompt with three good samples beats a clever prompt with none.
The second thing guides miss is that the biggest gain is not quality but speed. A good-enough reply in 20 minutes beats a perfect reply the next morning, because the prospect has usually contacted two other firms. If AI drafting does nothing except get your first reply out faster, it has paid for itself.
Related reading
For a tool-focused view, see AI email writer. For editing AI drafts so they read naturally, AI writing editor covers the editing pass. If most of your email goes out through workflows, the email writer guide also shows where AI drafting sits inside an automation.
Using AI to write the email you have not decided the purpose of. If you cannot state in one sentence what the reader should do after reading, the draft will wander. Write that sentence first and put it in the brief.
Frequently asked questions
Will prospects know the email was written by AI?
They will if it sounds like the average email. With your own samples, rules and the contact's real details, the draft reads as your firm wrote it, especially after a person adds one specific line.
Should AI send emails automatically?
Only for low-stakes, factual messages such as confirmations and reminders. Replies to enquiries, proposal follow-ups and anything to an existing client should be reviewed by a person before sending.
How many voice samples do I need?
Three to five good ones are enough to shift the tone. Pick emails that got replies, from the person whose writing your team most wants to match, and refresh them each quarter.
Can AI write our newsletter as well as one-to-one replies?
Yes, but treat it as a first draft. Give it the topics, the facts and your rules, then edit for accuracy and voice. A newsletter reaches your whole list, so the editing time is worth keeping.