How to write a cold email that gets replies (not just opens)
The anatomy of a reply-worthy cold email: a first line that proves homework, one problem, one proof point, one ask, and follow-up rules that don't burn goodwill.
By Norbelys Chirinos, Co-founder
Founder-reviewed ·How we research and correct articles
Opens are vanity — half of them are robots anyway. Replies are the game. And replies come from emails that read like a sharp colleague wrote them, not a sequence tool. Here’s the anatomy.
The subject line: small, lowercase, internal
The best cold subject lines look like they came from inside the company:
two to four words, lowercase, zero punctuation tricks. pipeline math,
re: your hiring page, quick one. You’re not trying to sell in the
subject — you’re trying to not look like marketing.
What kills you: ALL CAPS words, emoji, “{firstName}, exciting opportunity!”, and anything that smells like a newsletter.
Why lowercase and short work isn’t a style preference — it’s pattern matching. Spam filters and human skim-readers both learned the visual shape of marketing email: capitalized words, exclamation points, a subject long enough to need truncating on mobile. A subject that looks like something a coworker would type — because a coworker actually would type it that way — skips the pattern entirely. It’s the same reason checking your subject line length matters more than most senders assume: three words that fit fully on a phone screen read as a real message before the recipient has consciously decided anything.
The first line: proof of homework
The reader decides in one line whether this is mass spam or a message for them. So spend the line proving you did homework:
“Saw you just opened the Austin office — guessing outbound for the new team starts from zero.”
Not flattery (“Love what you’re building!”), not biography (“My name is X and I’m the founder of…”). An observation specific enough that it couldn’t be pasted into a thousand emails. If you use AI to draft these, our prompt library has openers built around exactly this rule.
The test that catches a weak first line before you send it: could this exact sentence be pasted into an email to a different company and still be technically true? If yes, it’s not personalization, it’s a mail-merge field wearing a personalization costume — and readers pattern-match that just as fast as they pattern-match a spammy subject line. Compare the two openings side by side:
| Weak (generic, passes as “personalized”) | Strong (specific, provably about them) |
|---|---|
| “Hope you’re having a great week running [Company]!” | “Saw [Company] just opened the Austin office — guessing outbound for the new team starts from zero.” |
| “I noticed you’re in the SaaS space.” | “Noticed your pricing page added a usage-based tier last month.” |
| “Companies like yours often struggle with X.” | “Your last three job posts are all for AEs — usually means SDR headcount hasn’t kept up.” |
The right column all share a trait: each sentence would be false, or at least odd, if sent to a random other company. That’s the actual bar.
The body: one problem, one proof, under 90 words
Cold emails die of obesity. The discipline:
- One problem, stated in the reader’s vocabulary — not your feature list.
- One proof point, concrete and unembellished: “we got a 9-person agency from 2% to 7% reply rate in six weeks” beats three paragraphs of adjectives.
- Under 90 words total. It gets read on a phone, between meetings, by someone who didn’t ask for it. Respect that.
The ask: answerable with one word
“Do you have 30 minutes Thursday?” is a calendar negotiation. “Worth a look?” is answerable from a phone in one second. Lower the activation energy: interest first, logistics after they say yes.
Five ways a technically good email still gets ignored
Nailing subject, first line and ask still isn’t a guarantee — a handful of quieter mistakes sink otherwise solid emails:
- Two asks instead of one. “Worth a look? Also, do you know who handles procurement?” forces a decision between two questions, and the easiest decision is to answer neither.
- The proof point is about you, not them. “We’ve helped 200 companies” is a claim about your business. “We got a 9-person agency from 2% to 7%” is a claim the reader can map onto their own situation — the second is the one that moves people.
- Sent at a dead time. Timing doesn’t rescue weak copy, but it can bury strong copy under a Monday-morning inbox flood — see the actual data on when to send before assuming your send time is neutral.
- A signature block longer than the email. Four lines of title, logo, socials and a legal disclaimer under a 60-word message reads as marketing regardless of how human the body sounded.
- No signal that a human wrote it. A closing line with zero personality — no first name, no natural phrasing, nothing that couldn’t have come from a template — undercuts every “sharp colleague” cue the rest of the email built.
Follow-ups: add, never bump
Most replies come from follow-ups — but only the ones that add something. The rules:
- Never send “just bumping this.” It converts goodwill into spam reports.
- Each follow-up brings one new thing: a different angle, a relevant resource, a sharper proof point.
- Three to four touches, then a graceful exit. The breakup email (“closing the loop — if X ever becomes a priority, this is where I am”) often out-replies everything before it.
- Stop instantly on reply. Nothing torches a warm conversation like a robot follow-up arriving after a human answered. (Norbelys stops sequences automatically the moment a real reply lands.)
- Space touches by days, not hours. Three to five business days between each follow-up gives a busy reader room to see and act on the previous one before the next arrives; stacking follow-ups daily reads as pressure, not persistence, and pressure is what gets reported.
Worth internalizing about the reply curve itself: in most sequences a meaningful share of total replies — often a third or more — land on the second or third touch, not the first. A single-send campaign isn’t testing whether cold email works for a prospect; it’s testing whether that prospect happened to be free the exact moment your one email arrived. The follow-up is where a genuinely interested-but-busy reader gets the second chance a single send never gives them.
Then: measure replies, not opens
Write two versions of your subject or first line, send both, and let real replies pick the winner — that’s the one metric machines can’t inflate. It’s how A/B testing works in Norbelys, and it’s how you get honestly better with every campaign, instead of optimizing for whatever Apple’s prefetcher liked this week.