How to Write Cold Emails with AI That Don't Sound Like AI
AI can write a cold email in seconds. The problem is that buyers can tell. The flattering opener, the three part structure and the "I'd love to connect" close all mark an email as machine written before the second sentence. This guide shows how to use AI to write cold emails that sound like a sharp human wrote them to one person, because that's what gets replies.
Why most AI cold emails fail
The typical AI cold email fails for predictable reasons:
- No real trigger. It opens with a compliment ("I was impressed by your company's growth") instead of a specific, recent event.
- Personalization without relevance. It mentions the prospect's alma mater or a LinkedIn post, then pivots to a pitch that has nothing to do with it.
- Too long. AI defaults to thorough. Buyers read on their phones between meetings.
- Recognizable phrasing. "I hope this finds you well." "I came across your profile." "Leverage." "Game changer." "Excited to."
- A vague ask. "Let me know if you'd be open to connecting" gives the buyer nothing to say yes to.
None of these are AI problems. They're instruction problems. Tell AI to "write a cold email" and it produces the average cold email, which is exactly the one buyers have learned to delete. New to this? How to Tell If Something Was Written by AI walks through each pattern on an annotated example.
What a good cold email looks like
Before prompting anything, know the target. A first touch email that earns replies is:
- Relevant: tied to a real trigger and a problem the buyer owns
- Short: 50 to 125 words
- Plain: reads like one person writing to another
- Specific proof: one concrete result from a similar customer
- One easy ask: a single question the buyer can answer in a few words
Relevance beats personalization. Mentioning that a prospect went to the same college as you is personalization. Connecting their 18 new prospecting job postings to the ramp time problem you solve is relevance. Only one of those gets meetings.
Step 1: Give AI a real reason to write
Every strong cold email starts with research: a trigger from the last 90 days and a hypothesis about the problem it creates. AI can find both quickly, but you need to verify them before they go in an email. (The full method is in How to Use AI for Account Research.)
Good triggers for cold email:
- A hiring surge in the team you support
- A new executive in your buyer's function
- A funding round, acquisition or expansion
- An earnings call comment about a problem you solve
- A product launch or move into a new market
If you can't find a trigger, that's useful information too. Put the account in a lighter touch tier rather than forcing a weak one.
Step 2: Train AI on your voice
AI will write in its own default voice unless you show it yours. Do this once, in your AI tool's Project or custom instructions:
Paste 3 to 5 of your best emails, the ones that got replies, and label them:
These are examples of my writing voice. Match their length, tone and structure. Don't copy their content.
Add a banned phrase list. Keep it short and focused on the worst offenders:
Never use: "I hope this finds you well," "I came across," "I wanted to reach out," "just circling back," "just checking in," "quick question," "leverage," "synergy," "game changer," "excited to," "I'd love to," "touch base."
Describe your style in plain terms: short sentences, no exclamation points, no buzzwords and one idea per sentence.
After a week of drafting, ask AI: "Based on how I've edited your drafts, what patterns should you learn?" Add its answer to your instructions. Your drafts will need less editing every week.
For a 4 week plan that turns this setup into a habit and checks it on 10 sent emails, see I want my emails to sound like me.
Step 3: Use a framework, not a vague request
AI follows structure much better than it follows adjectives. "Make it compelling" produces fluff. "Use this four part structure" produces a usable email. Three frameworks work well for cold email:
Scroll sideways to see the whole table.
| Framework | Structure | Best for |
|---|---|---|
| Trigger, Problem, Proof, Ask | What happened, the problem it usually causes, who you helped, a question | Accounts with a clear, recent trigger |
| Observation, Insight, Question | Something you noticed, a point of view, an open question | Senior buyers who respond to ideas |
| Problem, Cost, Path | A known pain, what it's costing them, a way out | Segments with well understood pains |
Name the framework in your prompt, and the email will follow it.
Step 4: The cold email prompt
Goal: book a 20 minute call with {name}, {title} at {company}.
Context: use my voice examples and context. Trigger: {trigger, with date}.
Inputs: {paste your verified research}. Use only these facts about {company}.
Rules:
- Framework: trigger, problem, proof, ask
- Under 90 words
- Open with their situation, not with us
- One proof point, taken only from my approved customer examples
- End with one question they can answer in a few words
- No banned phrases
Format: 3 subject lines under 5 words, then 2 versions: one direct, one curious.
An example
Here's the kind of output this produces, for a fictional prospect:
Subject: 18 sdrs at acme
Hi Dana,
Saw Acme has 18 prospecting roles open right now. When teams hire that fast, the bottleneck is usually ramp: new reps take months to book their first meetings.
We helped [Customer] cut prospecting ramp from 6 months to 4 with AI coaching built into onboarding.
How long is it taking your newest reps to book their first meeting?
Sam
That's 63 words. It's specific, it's about Dana's world and the question is easy to answer. It doesn't have to be clever. It has to be relevant.
Step 5: Make AI critique the email as the prospect
This is the step most reps skip, and it's often the one that improves an email the most. Ask AI to switch roles from writer to skeptical reader:
You are {title} at a {size} {industry} company. You get about 80 cold emails a day and delete most in 3 seconds. Read this email as that person. Score it from 1 to 10 for how likely you'd be to reply. Tell me: the exact moment you'd stop reading, anything that feels generic or automated, and what would make you reply. Then rewrite it to score at least 8.
Run it, revise and run it again. Two rounds usually move an email from forgettable to worth sending.
Step 6: The 60 second human edit
Never send an AI draft untouched. Every email gets this quick check:
- Verify every fact about the prospect against a source you opened yourself.
- Delete the first sentence if it's a warm up. The second sentence is usually the real opener.
- Read it aloud. Rewrite anything you'd never say to someone's face.
- Check the ask. One question, specific, easy to answer.
- The real test: if you received this, would you reply?
This takes a minute. Skipping it is how invented facts and robotic phrasing reach buyers.
Subject lines that work
Short, lowercase or sentence case, specific and never clickbait:
18 sdrs at acmeyour q2 pipeline commentramp time at acmeidea for the new team
Ask AI for 10 subject lines and pick one. Avoid anything that looks like marketing: no emojis, no ALL CAPS and no fake "Re:" prefixes.
Follow ups that add value
The follow up is where most sequences go generic. "Just bumping this to the top of your inbox" adds nothing. Every follow up should bring a new angle:
Scroll sideways to see the whole table.
| Touch | Angle |
|---|---|
| Email 2 | A relevant insight or data point from your industry (only numbers you can source) |
| Email 3 | A short story about a similar customer |
| Email 4 | A different pain the same persona usually has |
| Final email | A polite close that leaves the door open |
Write follow up email {number} for {name}. The first email is below. This one must use a completely different angle: {angle}. Under 80 words. No reference to "following up" or "bumping." One easy question at the end.
For an angle bank, post meeting follow ups and how to revive stalled threads, see How to Write Follow Up Emails with AI That Add Value.
For a full multichannel sequence prompt with email, phone and LinkedIn, see 15 AI Prompts for Sales. For the LinkedIn touches in that sequence, including connection notes, comments and InMail, see How to Write LinkedIn Messages with AI.
Personalization at scale: the tier system
You can't write a fully custom email for every prospect, and you shouldn't try. Match effort to account value:
Scroll sideways to see the whole table.
| Tier | Approach | Time per prospect |
|---|---|---|
| Top priority | Fully custom: researched trigger, tailored proof, critiqued draft | 10 to 15 min |
| Second priority | Persona and industry template with a custom first line and trigger | 3 to 5 min |
| Third priority | Segment template with light personalization | Under 1 min |
AI makes top priority faster and makes second priority feel closer to top priority. It doesn't make it safe to skip human review at any tier.
The deliverability warning
AI makes it easy to write a lot of email. Writing more is not the same as writing better, and volume has real costs:
- Domain reputation: sending large volumes of low engagement email can push your messages into spam folders, for everyone on your domain.
- Brand damage: one flawed template sent to 500 people is 500 bad first impressions.
- Compliance: CAN SPAM, GDPR and similar laws apply to AI written email exactly as they apply to any other email. Honor unsubscribes and follow your company's sending rules.
The rule: never auto send AI copy nobody reviewed. Automation should prepare drafts for you, not send them for you. (More on where automation helps and hurts in AI Sales Automation: What to Automate and What Not To.)
Handling replies with AI
AI is also useful for responding to the replies you get. Paste the reply and your thread:
A prospect replied: "{reply}". Write a response that respects their answer and gives them one easy reason to keep talking. Under 60 words. No pressure, no guilt. If the right move is to accept and set a follow up date, write that instead.
The most common replies to prepare for: "Not interested," "Send me some info," "Talk to [someone else]" and "Try me next quarter."
Frequently asked questions
Can buyers tell if a cold email was written by AI?
They can tell when an email is generic, and generic is what AI produces by default. An email built on a real trigger, written in your voice and edited by you reads as human, because a human made the decisions that matter.
What's the best AI tool for writing cold emails?
Claude, ChatGPT and Gemini all write strong emails when given good instructions. The difference comes from your voice examples, your research and your editing, not the tool.
How long should a cold email be?
For a first touch, 50 to 125 words. Under 90 is a good target. If you can't explain the reason for your email in that space, the problem is usually the angle, not the length.
Should I use AI to personalize emails at scale?
Use it to make each tier faster, not to fake personalization. A true trigger and an honest hypothesis about the buyer's problem work better than a compliment about their LinkedIn post, at any volume.
Is it legal to send AI written cold emails?
The same laws apply as for any cold email. In the US that includes CAN SPAM. In Europe and the UK, GDPR and related rules apply. Follow your company's sending policies and always honor opt outs.
Write better emails, faster
The Wicked Growth Club AI for Sales Certification spends a full session on AI outreach: voice training, frameworks, sequences and the edit habits that keep emails human. Reps leave with a ready to send sequence for one of their own top priority accounts.
Start with the fundamentals in the practical guide to AI for sales.
AI for Sales Certification
Practice this on your own accounts
10 self paced sessions, an optional weekly live class and a graded capstone, all practiced on your own accounts. Membership is $50 a month. Add the $250 certification when you're ready.
Not sure where to start? Take the 2 minute assessment