How to Write Cold Emails With AI That Get Replies
To write cold emails with AI, do the research yourself and let the AI do the drafting. Give it one specific fact about each prospect, the problem you solve and the single thing you want them to do, then ask for a short email built around that fact. Keep it under about 120 words with one ask, cut every line that sounds generated, and read it as the prospect before it goes out.
The AI is fast at the parts that don’t matter much, like phrasing, variations and subject options. It’s bad at the part that matters most: knowing why this person should care today. That’s the split this guide is built on.
Why do most AI cold emails get ignored?
Because they read like every other AI cold email the prospect got that week. The pattern is easy to spot:
- An opener that compliments something vague: “I came across your profile and was impressed by your work.”
- A paragraph about the sender before the reader’s problem shows up.
- Three benefits in a row, each one a bit grander than the last.
- A closing ask for “a quick 15-minute call” with no reason to take it.
None of that is the model’s fault. It writes the average of what it has seen, and the average cold email is bad. A better prompt helps a little. Better inputs help a lot.
What should you give the AI before it writes?
Treat the AI like a new hire writing their first outreach. It needs a brief, not a vibe.
| Input | Example | Why it matters |
|---|---|---|
| One specific fact about the prospect | “They just opened a second warehouse in Ohio” | Gives the email a reason to exist |
| The problem you solve, in their words | “Inventory counts drift between two locations” | Keeps the email about them |
| Your proof, if you have it | “[a client result you can name]” | Makes the claim believable |
| One ask | “Reply with who handles inventory” | Gives them an easy next step |
| Your voice | Two or three emails you wrote yourself | Stops the default AI tone |
If you don’t have proof yet, leave a [placeholder] and don’t let the AI fill it in. An invented case study in a cold email is the fastest way to lose a prospect who checks.
How do you structure a cold email?
Four parts, in this order, and nothing else:
- The trigger. One line on the specific fact: what you noticed and where.
- The problem. One or two lines on what that fact usually causes, said the way they’d say it.
- The offer. One line on what you do about it, with proof if you have it.
- The ask. One small, specific next step.
That shape fits before-after-bridge closely: the trigger and problem are the before, the offer is the bridge. If you want a longer structure for a follow-up, see the copywriting frameworks guide.
A small ask gets more replies than a big one. “Worth a look?” or “Who handles this on your team?” is easier to answer than a calendar link to a stranger.
How long should a cold email be?
Short enough to read on a phone without scrolling. Around 50 to 120 words covers most cold emails. Every extra sentence is one more place for the reader to decide it’s not for them.
Ask the AI for a word limit in the prompt, then cut further by hand. The first draft almost always has a sentence that restates the one before it.
What should the subject line say?
What the email is about, in plain words, with the trigger if it fits. “Second warehouse in Ohio” beats “Quick question” because it tells them this isn’t a mass send. Our guide to email subject lines covers length and the patterns that work.
How do you make AI cold emails sound human?
Edit for the tells. Read the draft out loud and cut:
- Openers about yourself or about “reaching out”.
- Words like “leverage”, “streamline”, “elevate” and “unlock”.
- Lists of three where two would do.
- Sentences that all have the same length.
- Any compliment you couldn’t back with a specific.
Then put one sentence in that only you would write, like a detail from your own work or a plain opinion. Making AI writing sound human and editing AI drafts go further.
What rules do cold emails have to follow?
Rules depend on where you and your prospects are, so check the ones that apply to you. In the US, the FTC’s CAN-SPAM compliance guide covers commercial email, including an honest subject line, a valid physical address and a way to opt out. If you send in volume to Gmail users, Google’s email sender guidelines also matter for whether your email reaches the inbox at all.
Can a tool check a cold email before it goes out?
You can do it by hand with the checklist above. A tool helps when an AI is drafting many emails and they start to sound alike.
Copix plugs into Claude, ChatGPT and Cursor and gives the AI a review tool. For an email it reads the body with its subject and preview text and returns a verdict with findings per field: length, links, unproven claims and the AI tells, each with a fix. In the same pass its voice check flags generated-sounding lines, and its persuasion check catches a second call to action competing with your ask. See the tools and the plans.
FAQ
Can ChatGPT or Claude write cold emails?
Yes, they draft well when you give them the research, the problem and one ask. Without that, they write generic emails that look like everyone else’s. Always read and edit before sending.
Should I personalize every cold email?
Personalize the first line with something specific to that prospect. Keep the rest of the email the same for everyone with the same problem. One true specific does more than three fake compliments.
How many follow-ups should I send?
A few, spaced out, each with a new angle or detail rather than “just bumping this”. Stop when someone asks you to, and honour opt-outs immediately.
Is it legal to send cold emails?
In many places it is, with conditions. In the US, CAN-SPAM sets rules for commercial email. Other countries have stricter consent rules, so check the law where your recipients are.
What’s the best call to action for a cold email?
A small ask that’s easy to answer, such as a yes or no question or a request to point you to the right person. Save the meeting request for after they reply.