How to write a cold email that gets replies in 2026
Cold email fails twice: the provider blocks it, then the copy asks too much. The deliverability checks, the four lines and the maths that move reply rate.
Most cold emails fail twice. They fail technically, because the mailbox provider never lets them land, and they fail commercially, because the copy asks a stranger for something they have no reason to give. I run B2B outbound through Sail Force Marketing across Europe, Africa and the US, and I have never seen a reply-rate problem that was really a writing problem alone. Fix both layers, in the order they break.
Check delivery before judging the writing
Make sure the message reaches the inbox before you interpret reply rates. Authentication, complaint rates and opt-outs matter, but they deserve their own checklist rather than a second copy of one here. Use the cold email deliverability guide for those checks, then return to the four-line writing method below.
One email, one idea, one ask
The most common structural error I see is a cold email trying to do four jobs: introduce the company, prove the category, list three use cases and book a call. Each addition lowers the chance of the one outcome you want. A first email has exactly one job, which is to earn a reply. Not a meeting, not a demo, a reply.
A cold email is not a pitch. It is a request for permission to have a conversation, so write it as the smallest credible request you can make.
The four lines that earn a reply
The subject line
Three to five words, lower case, no punctuation tricks, no curiosity bait. It should read like internal mail from a colleague, because that is the standard the inbox judges it by. "question on your Q4 onboarding" works. "Unlock 3x Pipeline Growth!" does not, and it teaches the filter that your domain sends marketing.
The opening line
One sentence that could only have been written to this person. Not their job title, not their funding round, not the city they are in. Something they said, shipped or published. If you cannot write that line in under a minute, the contact is not researched enough to send to, and this is where most personalisation efforts quietly become mail-merge.
The claim
One sentence naming the problem in their language and the outcome you move. Name a mechanism, not an adjective: "we replace the manual credit check your ops team runs before each shipment" beats "we help companies streamline operations". If you have a published result, use it. If you do not, say what you believe and why, which is more credible than a borrowed statistic.
The ask
One question that is cheap to answer. "Is the manual credit check still done by your ops team, or has that moved?" invites a one-word reply and gives you qualification either way. "Do you have 30 minutes on Thursday?" asks a stranger for the most expensive thing they own.
Assembled, it looks like this, and it should stay under 90 words:
- Subject: question on your [specific process]
- Line 1: Saw your [talk, post, release] on [specific thing]. The part on [detail] is the reason I am writing.
- Line 2: We [mechanism] for [their type of organisation], which usually shows up as [the number it moves].
- Line 3: Is [specific process] still handled by [team], or has that changed this year?
The arithmetic of outbound
Reply rate is a vanity metric on its own, because "no thanks" is a reply. Track positive reply rate, meaning replies that agree to a conversation, and work backwards from the revenue you need. The maths is simple enough to do on paper, and it is the fastest way to see whether you have a volume problem or a message problem.
Suppose you need four new opportunities a month. At a 3% positive reply rate and a 60% show rate, you need roughly 220 well-researched contacts a month to produce those four. Double the positive reply rate to 6% and the same four opportunities cost you 110 contacts, half the research and half the domain risk. That is why I prioritise reply quality over list size every time. The lever is the message, and it compounds.
Volume hides a broken message. Fix the reply rate at 100 sends before you pay for 10,000.
The four numbers I keep on one line in a spreadsheet: contacts sent, positive reply rate, meetings held, and spam rate. If spam rate is climbing, stop sending and fix the list, whatever the pipeline pressure. A burnt domain costs months.
Six failure modes, and what to do instead
- Buying a list. Bounce rates and complaints follow immediately. Build the list yourself against a written definition of who qualifies, which is the bar I set out in the qualification bar that decides whether your pipeline is real.
- Sending from your main domain. Use a dedicated outbound domain, warmed over two to three weeks before volume.
- Personalising the wrong field. A merged company name is not personalisation. One researched sentence is.
- Asking for time in email one. Ask a question first. Meetings come from replies, not from calendar links.
- Writing to the buyer you wish existed. Write to the person whose week is worse because the problem exists.
- Following up by nagging. "Bumping this to the top of your inbox" adds nothing. Each follow-up should carry new information.
Give each follow-up a different job
The first email starts a conversation; later messages should add evidence, a new angle or something useful. Do not send the same meeting request in different words. For the complete cadence and scripts, use the B2B follow-up sequence guide. Stop immediately when someone opts out or gives a clear no.
What to do this week
Pick 25 contacts you can each research in under a minute. Check SPF, DKIM and DMARC on your sending domain before the first send, connect Postmaster Tools, then write the four lines for all 25 by hand with no template. Count positive replies at the end of the week. If it is under two, the message is wrong and no amount of volume will rescue it. Rewrite the claim and the ask, send 25 more, and only scale once the arithmetic works. More on this in the Sales Growth library.