Cold email copywriting for founders
Cold email copywriting is the craft of writing a message a stranger can answer in one line. It has two jobs: make the reason you are writing legible up front, and make the ask small enough to answer without a decision. Subject line, structure, proof, sign-off: all serve those two jobs, or they are decoration.
By Rahul Bageria, co-founder · Updated August 2026 · 25 min read
Written by operators who run outbound for seed-stage teams, not by a company selling you a sending tool.
What cold email copywriting actually is
Almost every page on this subject teaches marketing copywriting with the word cold in front of it. That is the wrong parent discipline, and it is why so much of the advice does not survive an inbox.
Cold email copywriting is writing to someone who did not ask, in a channel where ignoring you is correct. The job is not to convert attention. It is to be worth four seconds.
Somebody who already handed over their attention. They clicked the ad, opened the page, bought the magazine. David Ogilvy's line about the headline being eighty cents of your dollar is a fact about that person.
Somebody triaging an inbox they never asked you to enter, on a phone, between two other things. Nothing was chosen. The transplanted rules break at exactly that join.
Marketing copy converts attention you already have. Cold copy has to be worth the attention it takes. That is why a clever headline is rewarded on a landing page and punished in an inbox.
Where the craft sits
Three things decide whether outbound works, and they fail separately. Copy is the only one a founder can fix this afternoon, which is why it collects all of the attention and most of the blame.
| The part | What it decides | Taught on this page |
|---|---|---|
| The offer | Whether anybody would buy this at all. Get it wrong and no sentence rescues you. | No. Fix it before you touch a word of the email. |
| The copy | Whether they answer this one. The reason, the claim, the ask, the length. | Yes. All of it, and nothing else. |
| The sequence | How many chances you get, and how far apart you take them. | No. A different problem with different failure modes. |
Copy is a multiplier on a real reason and it cannot manufacture one. When the reason is missing, the fix is upstream in who you write to and when, which is what signal-based selling is for.
The two ways a cold email dies
Every cold email that fails does so in one of two ways, and the two need completely different fixes. Confusing them is why founders spend their evenings rewriting the wrong email.
The email is polite, grammatical, correctly spelled, personalized with a token, and it has no reason to exist.
- ✕It survives the deletion test intact
- ✕Nothing happened that entitled you to write
- ✕Rewriting produces a better-written nothing
- ✕The fix is upstream, in targeting and timing
The reason is real, the reader half agrees, and the email still dies because answering it is more work than ignoring it.
- !The ask needs a calendar and a decision
- !Or there are two asks wearing one question mark
- !Or a claim has to be evaluated before anything else
- !The fix is entirely in the craft, and it is fast
The two ways a cold email dies are having nothing to say, and giving them no way to say yes. The first is a targeting problem wearing a writing problem's clothes, and better sentences do nothing for it. The second is the one this page fixes, and it is the faster win of the two by a distance.
Relevance is not personalization
The whole category uses the two words interchangeably. They are different things, they are produced in different ways, and only one of them still works.
Relevance is about the reader's situation: what changed, what it costs them, what they are now on the hook for. Personalization is about the reader. One can be generated. The other has to be found.
Open on their trivia
I came across your profile and loved your recent post on RevOps hiring.
- ✕A sentence about your afternoon
- ✕Any tool writes it in a second
- ✕Nothing in it needed this person
Open on their situation
You are the first RevOps hire, and the posting puts CRM consolidation in your first 90 days.
- ✓True of this account and few others
- ✓Names what they are on the hook for
- ✓Cannot be generated cheaply
The check takes five seconds. Delete the first name, the company name and the sign-off. If it still makes sense sent to anybody, you wrote a brochure with a merge field. When drafts keep surviving that test, the problem is not the prose: nothing at those accounts is worth writing about. Our signal library catalogs the events that are.
The five parts, and what each is for
Jeb Blount named the four elements of a prospecting email in 2015: hook, relate, bridge, ask. The structure holds. The weighting does not, because his hook was written for an inbox receiving a fraction of today's volume.
A label, not a hook
Six words or fewer, flat, describing what is inside. It exists so the message looks like correspondence rather than a campaign. Clever costs more than it earns here, and you have no way to tell whether it worked.
The only line they will definitely read
It answers one question: why is this person writing to me, now. A compliment answers nothing. A flat statement of the thing you observed, with no adjectives hung on it, answers it completely.
The part that carries the reason
What the thing you observed actually costs them, in their terms rather than yours. This is the load-bearing part now, and the part templates delete, because it is the only part nobody can write in advance.
One claim, and only one
How you help, stated as a mechanism a stranger can follow without trusting you. Not a case study, not a multiple, not a wall of logos. One sentence, checkable against what the reader already knows.
The smallest question that moves this on
Something answerable in one word, from a phone, without opening a calendar. If answering requires a decision it gets deferred, and deferred is how almost every cold email actually dies.
The opener has one job
Rapport is not the opener's job. Attempting rapport is how most first sentences waste the only line the reader is guaranteed to see.
A sentence about the sender's afternoon. It reports that you did some browsing, which was never news to anybody, and it burns the one line they are certain to read.
A compliment asks the reader to accept flattery from a stranger before they know what is wanted. That is a small debt nobody agreed to take on.
Introduces the sender, which the reader can already see in the From line. It spends the guaranteed sentence on information they had before they opened it.
What works instead is usually a flat statement of the thing you observed, with nothing decorative on it, so the reader can check it against their own week and get to the second line.
The opener I trust is the one I could not have written last week. If the same first line would have worked a month ago, it is not an opener, it is a greeting, and the reader has already had forty of those today.
The opener also has a clock on it. The event you noticed stops being news, and the same sentence sent six weeks late reads as scraping rather than attention. Windows differ by event type, which is a separate question from how you write the line.
The claim you are entitled to make
You get one claim in a cold email. Make it a mechanism rather than a multiple, and the reason has nothing to do with modesty.
Numbers feel like proof
The reflex is to lead with the strongest result you have, because a number looks like evidence and evidence looks persuasive. In a first cold email it is neither, and it quietly costs you meetings.
Gong Labs measured the cost
Across 132,552 cold emails, leading with ROI language, the word itself, a multiplier such as 2x, or a percentage, cut success by 15%. Success meant a meeting booked in ten days. Vendor data, sample and definition both disclosed.
A number is homework
A figure from a stranger is a claim the reader has to evaluate before they can use it. Evaluating is work, and work at the top of an unsolicited email is what the delete key is for.
Say how it works instead
A mechanism can be understood without being believed. "We do the deduplication before the migration" is checkable against the reader's own knowledge in about a second. "Three times faster" is not checkable at all.
When a number does belong
Once the reader has a reason to care and a mechanism to attach it to, a number is useful and often necessary. That moment is the second email or the call, not line four of the first message.
The ask
Ask for interest, not for time. The difference is friction, and friction decides everything that happens in the last sentence of a cold email.
Gong Labs' analysis of 304,174 emails found interest-based asks beat both specific-time asks and open-ended ones on cold email, with success defined as a meeting booked within ten days. Vendor data again, sample and definition disclosed. The ranking flips once a deal is live, which is why experienced reps get this backwards on cold.
| The ask | What it costs the reader | Where it belongs |
|---|---|---|
| Want the one-page version? | One word, typed with a thumb | The first cold email |
| Worth me showing you how the dedupe runs? | One word, plus mild curiosity | The first cold email |
| Open to a chat about this? | A yes that commits to something unspecified | Second touch, at a push |
| Do you have 15 minutes Tuesday? | Opening a calendar and making a decision | A live deal, never a cold one |
| Open to a call, or should I send some info? | Two decisions inside one question mark | Nowhere |
One ask. Not one ask plus a soft second one. "Happy to work around your schedule" is a third, and by then the reader is doing project management on your behalf.
Length, and why every rule you have read is wrong
Every length rule in this category traces back to datasets that were not cold email. Nobody says so, which is how folklore travels. Here is the chain, one link at a time.
Fifty to 125 words
It is on nearly every page that ranks for this query, usually with no source attached and occasionally with a study named but never described. It sounds precise enough that almost nobody checks it.
Boomerang, February 2016
Forty million emails, analyzed for reply rate against word count, reading level, question count and subject length. Real research, published openly, and still being quoted a decade later by people who have not opened it.
Not cold email
The population is emails whose sender set a reminder to follow up if nobody replied. That is business correspondence the sender expected an answer to: warm, often internal, largely between people who already knew each other.
Response rates above 50%
The study reports reply rates over 50% right across the 50-to-125 band. No cold outbound program on earth reaches that. The number is itself the proof that the sample is not your sample.
Same post, same problem
Write at a third-grade reading level, and ask one to three questions, come from that same 2016 post and inherit the same flaw. Both instincts are right. The evidence behind them is borrowed from strangers.
Four rules, four answers
Adestra said 90 characters and over. Marketo said four words. Boomerang said three to four. MailChimp found no correlation at all. All four measured opt-in marketing email, and measured it on open rate.
Write one observation, one claim and one ask, and the length takes care of itself. It usually lands between 50 and 120 words, which is why the folklore looks right. The floor matters more than the ceiling: under about forty words most cold emails are not concise, they are empty.
Links, formatting, and the spam-word myth
There is no list of spam trigger words. No mailbox provider publishes one. The genre traces to a HubSpot post listing roughly 400, which HubSpot has since walked back; the lists still circulating are copies of copies.
Gmail's sender guidelines, checked August 2026, cover authentication, a spam complaint rate under 0.30%, and one-click unsubscribe above 5,000 messages a day. On the content of your message they say exactly two things: links should be visible and easy to understand, and do not use HTML or CSS to hide content. That is the whole of it.
So stop rewriting "free" as "complimentary". What a writer genuinely owes deliverability is smaller and duller than the folklore, and it is four habits rather than a vocabulary.
A signature image is a payload the reader did not ask for and a legibility problem on a phone. Type your name.
Google's own guidance is that a link should be visible and easy to understand. A tracked link wearing a normal link's clothes fails that plainly.
Not because a link is a spam word. Because a link the reader has no reason to click is a small invitation to hit the spam button, and the complaint rate is a real threshold.
No bold-italic stacks, no nested bullets. Your reader is skimming on a phone, and so, now, is the summarizer sitting in front of them.
Scope note. This page owns the words. The one-time build of domains and mailboxes belongs to the infrastructure guide, and the science of getting delivered to the deliverability guide. The ranked causes of landing in the spam folder are in why cold emails go to spam, and the bulk-sender specification, one-click unsubscribe included, is in the sender requirements breakdown.
The five-pass edit
You cannot measure your way to good cold email copy at seed volume, so judgment has to become a procedure. This is ours: five passes, fixed order, about five minutes an email.
-
1
Delete
Take out the first name, the company name and the sign-off, then read what is left. If it still makes sense, it could go to four hundred people and no rewrite will save it. Stop editing and go and find a reason to write.
-
2
Claim
Count the things the reader has to evaluate before they can act. One claim, stated as a mechanism, is the budget. Two numbers and a customer story is not a stronger email, it is a longer queue of homework attached to a request nobody made.
-
3
Ask
Read the last sentence on its own. Can it be answered in one word, from a phone, without opening a calendar? If a second ask is hiding behind an "or", delete it. The second ask never wins, and it takes the first one down with it.
-
4
Cut
Read it out loud, which is the only way to hear the padding. Out goes every word doing no work, every long word standing in for a short one, every adverb its verb already carried. Length is what is left over. It was never the target.
-
5
Summary
Write the one-line summary a mail client would show above your email. If it comes out as "a vendor wants a meeting", the email has already failed whatever the prose is like. If it comes out as "someone noticed the CRM merge and offered to handle the dedupe first", send it.
William Zinsser put it as well as anyone in On Writing Well: "the secret of good writing is to strip every sentence to its cleanest components." The reason is not aesthetic. Daniel Oppenheimer's 2006 experiments found that needlessly complex language made writers look less intelligent, not more, whatever the quality of the argument underneath.
Nobody else runs the fifth pass. Since Apple Mail started summarizing and Gmail followed, the first reader of your first line is often a model, not a person, and the summary is what the human sees.
Want a second pair of eyes on the emails you are already sending?
Book a Fit CheckGood next to what only sounds good
Both emails below are ours, written for this guide. The company and the people are invented. This is not a client result and there is no reply rate attached to it.
Reads well, says nothing
Subject: Quick question, Maya
Hi Maya,
I came across your profile and was really impressed by your work at Northwind. Congrats on the new role, that is a big move.
I am reaching out because we help fast-growing B2B companies like yours modernize their revenue operations and get more value out of their data across the funnel. Our clients typically see a 3x improvement in data accuracy and save 10+ hours a week on manual cleanup.
Would you be open to a quick 15-minute call next week to explore whether there is a fit? Happy to work around your schedule.
Best,
Sam
- ✕Survives the deletion test untouched
- ✕Two numbers a stranger cannot check
- ✕Three asks and a calendar
Says one thing, asks one thing
Subject: the two CRMs from the Fairwater deal
Hi Maya,
You are Northwind's first RevOps hire, and the role posting puts CRM consolidation after the Fairwater acquisition in your first 90 days.
The part that eats the first month is not the migration. It is that the two systems define an account differently, so the first merged report gets argued with instead of used.
We do the deduplication and the field mapping before the migration, so the merged reporting survives its first board meeting. Two weeks of work, not a platform.
Want the one-page version of how it works?
Sam
- ✓Collapses without the two names
- ✓One claim, and it is a mechanism
- ✓Answerable with the word yes
| Pass | The first email | The second email |
|---|---|---|
| Delete | Remove Maya and Northwind and nothing changes. It could go to four hundred people. | Remove Northwind and Fairwater and the email collapses. That is the test working. |
| Claim | Two unverifiable numbers, both leading the paragraph, exactly the ROI shape Gong measured. | One claim, and it is a mechanism. There is nothing in it to disbelieve. |
| Ask | Three: be open to a call, explore whether there is a fit, work around a schedule. | One, answerable with the word yes, from a phone, with no calendar involved. |
| Cut | Really impressed, fast-growing, modernize, get more value out of, quick. Ninety-four words in the body and barely a dozen of them about Maya. | Eighty-nine words in the body, three short paragraphs, and no adjective doing decorative work. |
| Summary | A summarizer renders it as "a vendor wants a 15-minute call". That is the email. | "Someone noticed the CRM merge and offered to handle the deduplication first." |
The second email is not a perfect one and should not read like one. If it read like a template, it would have failed the point of the section.
What changed in the last 18 months
Most of this page is craft that holds. This section is weather, and it is dated on purpose. As of August 2026, four pieces of standard advice have quietly expired.
| What you were told | Where it stands, August 2026 |
|---|---|
| Test your subject lines and watch your open rate climb | Open rate has counted machines since Apple Mail Privacy Protection shipped in September 2021. Those tests have been measuring proxies for five years. |
| Personalize the first line with something from their profile | Generating a plausible line about anybody now takes no effort, so it signals none. Token personalization reads as automation because it usually is automation. |
| Your preheader is the second thing they read | Apple Mail has shown AI summaries in place of preheader text since iOS 18. Litmus, writing in June 2024, concluded that senders cannot control the summary and the opening carries the weight. |
| Gmail is a mail client | On 8 January 2026 Google announced Gemini 3 in Gmail: overviews that synthesize threads, free to all users, and an AI inbox view organized around summaries and to-dos. |
None of that changes what a good cold email is, only who reads it first. Your opening two sentences now feed a summarizer, and a growing share of readers meets the summary before they meet you. Bury the reason in sentence four and the summary is a sales pitch. State it in sentence one and the summary is the reason.
What is not new, despite the framing: short emails, one ask, plain language and a real reason were all correct in 2015 and are correct now for the same reasons. Anyone selling you a 2026 framework for those four things is selling you 2015 in a new hat.
Why you cannot A/B test your way to good copy
The standard advice is to test everything. At founder volume the arithmetic says you cannot, and the arithmetic is short enough to check rather than take on trust.
sends per variant, to detect a 50% lift on a 3% reply rate.
Standard sample-size arithmetic: enough sends that a real doubling shows up four times in five and a fluke fools you about once in twenty. Detecting a doubling, 3% to 6%, needs about 750 per variant. Reproduce it yourself.
So a founder sending 300 emails a week can detect a doubling in about five weeks, and never a 50% improvement before the market moves underneath them. Split that traffic across a subject-line test and a sign-off test and you are measuring noise with great confidence. This is not an argument against A/B testing. It is an argument for spending your sample on the one variable big enough to show up.
Test angles, not adjectives. An angle is a hypothesis about who your buyer is and what they are dealing with, and it moves things by multiples. An adjective moves things by an amount you will never see.
Writing for a segment, not a person
The field offers two options: spray a template, or hand-write every email. The unit that actually works sits between them, and almost nobody teaches it.
Twenty to sixty accounts that share one reason. Small enough that the reason is genuinely true of all of them, big enough that writing for it is worth an afternoon.
Two or three concrete details inside a fixed argument. The name of the acquisition, the line in the job posting, the tool they switched off last month.
When a segment replies you have found a group who agree with your read of their situation. That is a fact about your market, not just a good email.
The reason a segment shares is almost always an event: a round of funding, a new executive, a hiring push for the problem you sell against. Job changes are the one most founders can act on without buying anything, and post-funding outbound is the one where the whole crowd arrives on the same morning you do.
Segments are built rather than found: somebody has to decide which events matter for your ICP and which are noise, which is the job we call signal mapping.
What AI is genuinely good for here
Use a model to produce five angles. Do not use it to produce the finished email. The line between those two things is the whole of our position.
Generating options fast. Five different angles on the same segment in a minute, so you can pick the one you would not have thought of. Then throw its sentences away and write your own.
Deciding whether you are entitled to write at all. That judgment needs the reader's situation and an honest read of your own offer, and it is the only part of this that was ever hard.
The awkward part is that what models do well, producing fluent plausible sentences, is exactly what has been commoditised and devalued. A generated first line used to signal effort. Now it signals a tool. Which tool you pick matters less than where you stop using it, though they are uneven and we have compared the AI sales copywriting tools separately.
Every draft we ship is human-edited, without exception. Not as a principle. Because the pass a model cannot run is the one asking whether we were entitled to write this person this week, and that is the pass that decides.
Where the common advice is wrong
Most cold email advice is published by companies that sell sending tools, and it shows. Four claims that do not survive contact with a founder's actual week.
"Nail the subject line, personalize the first line, start from a proven template, and test your way up."
- ✕Spend your five minutes on the subject
- ✕Personalize from something on their profile
- ✕Start from a template that already works
- ✕Test subject lines against open rate
"Spend the five minutes on sentence one, find a reason worth writing about, and test angles rather than adjectives."
- ✓Sentence one now feeds the summary they see
- ✓Relevance survives automation, personalization does not
- ✓A template is a sentence with the reason removed
- ✓Open rate has counted machines since 2021
George Orwell got there first. In Politics and the English Language he pointed out that ready-made phrases will "construct your sentences for you." That is precisely what a cold email template does. The reason a templated email feels hollow is not that the reader has seen the words before. It is that nobody had to think to produce them.
Real science, and the most cargo-culted book in outbound. What the field took from it: fake scarcity, an unrequested audit, a logo wall. The principle that transfers is liking through similarity, and only when the similarity is real.
"Who owns this at your company" was novel when Predictable Revenue published it in 2011 and is now the most recognized shape in B2B outbound. A template every buyer can identify on sight is not a template, it is a signature.
The four elements still hold, but the hook was written for a 2015 inbox and cannot carry that weight now. Relate is the load-bearing part today, and it is the exact part every template deletes.
What to expect from better copy
Better copy is worth the afternoon and it has a hard ceiling. Both halves of that sentence matter, and the second half is the one nobody selling you a course will say out loud.
It converts the reasons you already have. Fewer emails die of confusion, more of the people who should reply do, and the replies you get are about the thing you wrote about rather than "who is this".
It cannot make a reason exist. On a wrong offer, a wrong list or a wrong moment, better sentences buy you a faster and politer no, delivered by somebody who was briefly paying attention.
Copy is a multiplier, and multipliers do nothing to zero. If ten emails you have already sent all survive the deletion test, the problem is who you are writing to and when, not how you are writing. Go and fix that instead. It is slower, less satisfying, and the only thing that works.
When that is the diagnosis, signal-based selling is the discipline that fixes it, and it carries the honest baseline for how little outreach of any kind gets an answer.
At low volume the most useful data is the replies themselves, including the rude ones. "We already use X" is a positioning problem. "Who is this?" is a relevance problem. "Not now" is a timing problem. Three different fixes, and no dashboard tells you which. Handling those replies well is its own job, and where hard-won pipeline quietly leaks away.
Key takeaways
- 1 Two jobs only: a legible reason, and an ask that needs no decision.
- 2 Delete the names. If the email survives, the writing is not the problem.
- 3 One claim, and make it a mechanism a stranger can check.
- 4 Ask for interest, not for time. Calendars are where cold emails die.
- 5 Five passes before you send: delete, claim, ask, cut, summary.
Questions founders ask
How long should a cold email be?
Should I personalize every cold email?
What should the call to action be?
Which words trigger spam filters?
Should I use AI to write my cold emails?
My copy is good and nobody replies. What now?
Co-founder of Real Good GTM. He has been a first business hire and Chief of Staff at seed-stage B2B startups, which meant writing the outbound himself and reading every reply that came back, rather than approving a deck about it. This guide is the part of the craft that survived that, and the parts of the standard advice that did not.
Connect on LinkedInThe three things next to this one
Copy is one of three variables. These cover the other two, plus the tools that draft the first version for you.
Cold email deliverability
Getting delivered is a different discipline from getting read: authentication, reputation, complaint rate, and what a sender genuinely controls.
Read the guideSignal-based selling
The worldview underneath this page: how a real reason to write gets found in the first place, and what to do when there is not one.
Read the guideBest cold email tools
The senders, warmup tools and inbox stacks compared honestly, including who is better off with a spreadsheet and one mailbox.
Compare the toolsWant us to read the emails you are sending?
Book a fit check. We'll look at your ICP, the reason you have for writing, and the emails already going out, then tell you which of the two failure modes you actually have. If copy is not your problem, we'll say that instead.
Book a Fit CheckNo hard sell. No fake numbers. Real good work speaks for itself.