How to write a cold email that gets replies (a 5-line framework)

A cold email that earns a reply runs about 50 to 125 words, opens on a line that could only have been written to one person, names one problem they already have, gives one reason to believe you can fix it, and closes with a single ask small enough to answer from a phone. Structure and specificity move reply rate by multiples. Sign-off wording, send time and formatting move it by a rounding error. Here is the frame, the job each line does, the test that tells you a line has failed, and one full rewrite scored against a real pre-send filter.

The five lines and the job each one has

Each line has exactly one job, and a line that does two jobs is doing neither.

1. Prove you looked. Fifteen words or fewer, unmistakably about them.
2. Name the problem in the words their team uses internally, not your category's words.
3. Bridge with one claim and one proof.
4. Make one ask, sized so the reply costs almost nothing.
5. Sign off. Your name. Nothing else.

The order is load-bearing. Lines one and two buy the right to line three; a message that opens on line three is asking somebody to care about your mechanism before they have agreed there is a problem.

Line one is evidence, and there is a mechanical test for it

The opener is not a greeting. It is evidence that a human spent time on this. The reader is deciding one thing in the first two seconds — is this a mail merge — and the only reliable tell is whether the first line could have been sent unchanged to four hundred other people.

The test. Paste your first line into a different prospect's email and read it. If it still reads fine, it is not an opener. It is filler, and the fix is to write one that breaks.

Works: *"You have three SDR roles open in Bengaluru and the specs list no dialer."* That sentence is false about almost everybody, which is precisely why it works.

Fails: *"I hope this email finds you well."* Fails: *"My name is Alex and I lead growth at Northwind."* The second one fails for a specific reason worth naming — it spends the highest-attention line in the message on you, and you have not earned that line yet.

There is a second-order failure that catches careful writers. An observation can be specific and still be a commodity: if everyone in their inbox this week is congratulating them on the same funding round, you have personalised nothing and the specificity is worthless. Find the thing nobody else bothered to look up. Worked personalisation examples by signal type sets out which signals are still scarce and which have been mined flat.

Line two states their problem in their vocabulary

Say the pain you believe they have using the words their team uses internally, not the words your category markets with. *"Reps rebuild the same list in a spreadsheet every Monday"* beats *"inefficient prospecting workflows"* — buyers do not search for your category name, they search for the thing that is annoying them.

The test. Could a person inside that company have said your sentence out loud, to a colleague, without translating it? If it needs translating, it is your language and not theirs.

One problem only. Naming three signals that you are guessing at all three, and the reader will pick whichever one is least true and use it to dismiss the rest.

Line three: one claim, one proof, and no feature list

Say how you help in a single sentence, then attach exactly one piece of evidence. Evidence means a named comparable customer, a mechanism specific enough to be falsifiable, or a number you can defend under a follow-up question.

If you cannot defend the number, describe the qualitative change instead. An invented statistic is the fastest way to lose a technical buyer permanently, and it is not recoverable — nobody re-opens their assessment of a vendor who was loose with a figure in a first email.

The test. Delete the sentence and ask whether anything the reader could disagree with has been lost. If not, you wrote a claim with no content: "leading", "best-in-class", "trusted by hundreds" and "end-to-end" all pass through the reader without registering.

Do not list features. The reader has not agreed there is a problem yet, so a feature list is an answer to a question nobody asked.

Line four: the ask sets the cost of replying

The ask moves reply rate more than any other line, because it sets what a reply costs. *"Worth 15 minutes next week?"* costs one word to answer. *"Book a demo"* costs a calendar decision, a diary check and an implied commitment to a process.

Give exactly one. Two asks in a cold email do not double your options — they force a decision the reader defers, and deferred is deleted. The most common failure in otherwise-good emails is a strong single ask followed by a softer second one added just in case. Cut the second one every time.

The test. Read your ask and count the words in the shortest honest reply. One or two words is right. If the shortest honest reply is a sentence, the ask is too expensive.

Line five: your name, and why the links belong in the reply

Your name. Optionally your company. No link, no attachment, no six-line signature with social icons.

There is a filtering reason as well as a psychological one, and it is worth being specific about the mechanics rather than hand-waving at "spam filters". Autocloz runs a deterministic pre-send scorer on every save and preview, and its link rule is: more than three links in the body adds four points per extra link, capped at twenty, with the resulting advice being to cut down to at most two. A known link-shortener domain — bit.ly, tinyurl.com, t.co, goo.gl or is.gd — adds a flat twelve on its own. A cold first touch has no prior engagement with that recipient to offset any of it.

On the first email, aim for zero links. The case study goes in the reply, where it is a response to interest rather than an unexplained attachment.

The subject line is a separate job with different constraints

The body earns the reply. The subject only earns the open, and it is optimised against a different set of rules.

• Thirty to sixty characters is the working range. Past roughly ninety it is clipped on mobile; past a hundred and twenty it is truncated nearly everywhere; under about eight it carries no context and reads as spam.
• Three to seven words, lowercase more often than not. Lowercase reads as a person typing rather than a campaign sending, and Autocloz's subject analyser deducts five points from a subject that is Title Case Across Every Word for exactly that reason.
• No trigger vocabulary. The subject-specific weights are heavier than the body ones: "act now" costs fourteen, "make money" and "earn $" cost eighteen each, and "you have won" costs thirty.
• No fake "Re:" or "Fw:" prefix. Both are penalised — six points and eight respectively — and the penalty is deliberately light because legitimate replies use them. It buys one open and costs the relationship.

More patterns with worked examples are in subject lines that get opened, and you can score a draft against these rules in the subject line tester.

One honest disagreement with our own tool, because pretending otherwise would be dishonest. The analyser deducts eight points from any subject with no merge variable, on the reasoning that cold opens benefit from a first name or a company name. For a first touch I think that heuristic is wrong: a subject containing the recipient's company name is the single most recognisable mail-merge signature in a business inbox, and the specificity belongs in line one where it can carry a real observation. Score your subject, then override that particular deduction on purpose.

One rewrite, scored against the real filter

Here is a bad email of the shape that ships every day — the offer is fine, the execution is not.

Subject: LIMITED TIME OFFER - free trial, act now

*Dear valued customer, Our platform helps companies of all sizes drive revenue growth with best-in-class analytics. There is no obligation and you can start a free trial today. Click here to book a demo, or reply and we will call now. Best, {{ first_name }}*

Run that through the shipped scorer, with five links in the HTML body, and the result is not marginal. Ten separate rules fire: the phrases "free" (5), "act now" (12), "limited time" (8), "click here" (8), "call now" (8), "no obligation" (8) and "dear valued customer" (10); three all-capital words at four points each (12); five links, two over the threshold (8); and an unresolved merge tag that reached the final text (25). The raw total is 104, clamped to 100 out of 100, severity "high" — a score the scorer's own documentation says should block a send. It returns four suggestions: rewrite the sales-y phrases, swap the capitals for sentence case, cut links to at most two, and check the merge-tag names against your lead fields.

Score the subject separately and it lands at 42 out of 100 at forty characters, losing eighteen for heavy capitals and thirty-two across three trigger phrases.

Now the same offer, same sender, same person, in 58 words.

Subject: your three open SDR roles

*Hi Priya,*

*You have three SDR roles open and the specs list no dialer.*

*Teams hiring that fast hit the same wall: reps ramp on email, then wait weeks for calling and LinkedIn to be provisioned separately.*

*We put all three behind one login, so a new rep dials on day one.*

*Worth 15 minutes next week?*

*Alex*

Body score: 0, severity "clean" — no rule fires at all. Subject score: 92 out of 100 at twenty-five characters, losing only the eight points for having no merge variable, which is the deduction argued with above.

The rewrite changed nothing about the product. It changed who the first line was about, cut three asks to one, replaced four adjectives with one falsifiable mechanism, removed every link, and deleted 55 words. That is the entire method, and you can run the same check on your own draft in the cold email spam checker before you send it rather than after.

Autocloz's free plan covers 5 users and 10 mailboxes, with the warmup ramp and SPF, DKIM and DMARC monitoring included — start free and score a real draft against your own list.

Five ways this framework still fails

The list is wrong. No sentence structure rescues a message sent to somebody who does not have the problem. A tight list at a modest reply rate beats a list ten times the size at a fifth of it, and costs less to send.

The timing is wrong rather than the email. Most positive replies arrive on the third touch or later. A perfect first email with no follow-up under-performs a decent one inside a five-touch sequence; how many follow-ups to send works through where the curve actually flattens.

It never reached a human. Copy cannot fix authentication. SPF, DKIM and DMARC have to be right, and a new mailbox needs a ramp before it carries campaign volume — Autocloz runs a 21-day smoothstep ramp on every new mailbox and counts warmup traffic against the same daily ceiling as campaigns, so a warming mailbox cannot quietly be double-loaded by a campaign launch.

You personalised the wrong thing. A commodity observation is not personalisation. If three other vendors cited the same public event this week, you have spent your best line on something the reader has already discounted.

You asked twice. Covered above and worth repeating because it is the most common flaw in emails that are otherwise good.

What a better email cannot fix, and what to measure instead

Four limits, stated plainly.

Copy cannot fix delivery. If messages are landing in spam, every rewrite is invisible. Diagnose in delivery order — authentication, then placement, then copy — because a copy test run on mail that never arrived produces a confident, wrong conclusion.

Copy cannot fix a wrong offer. If the problem you name is real and your solution is not credible for it, the email is doing its job and the answer is still no.

Open rate cannot judge your subject line. Apple's Mail Privacy Protection pre-fetches tracking pixels, so a meaningful share of recorded opens are machines rather than readers. Treat open rate as a smoke alarm for delivery, not as a performance metric, and judge subject tests on replies.

And nothing here promises a number. Reply rates vary by list quality, offer, market and season far more than by phrasing, and any page quoting you an expected percentage for your situation is guessing. What the framework does is remove the failures that are unambiguously costing you replies — a merged first line, three asks, five links, a subject in capitals — and leave the outcome to the offer and the list, which is where it belongs.

If your motion needs the same discipline applied across more than email, the honest comparison of a multi-channel sequencer against a dedicated sender is in the Autocloz and lemlist comparison, which sets out where each shape is the better answer.