How many follow-ups a cold email actually earns

You have the sequence half built and five tabs open, and every tab gives you a different number. Three days. Two days. Three to four business days. One week. Stop at four emails, or five, or seven.

Here is the answer you came for, and then the reason the tabs disagree.

Send four or five emails in total, spaced three to four days apart, and read the result on meetings booked rather than on replies. Four to five is where the two largest published datasets overlap: Instantly advises at least four, Belkins puts the practical range at three to five steps. Three to four days is the only whole-sequence spacing figure any of them publishes, and it is Instantly's.

The two datasets that measured this, and what each one counted

Instantly's 2026 benchmark report states it plainly: "58% of all replies are generated from step one in a cold email campaign", with the remaining 42% spread across every follow-up. The report covers January 1 to December 18, 2025, and describes its base as "billions of cold email interactions across thousands of active workspaces". No email count is given.

Belkins published the inverse in the same year, on a smaller and much more specific book of data: "steps 2 through 6 together account for 58.6% of all replies, whereas the initial email captures just 41.4%."

Same two digits, opposite conclusions, and both are true. The difference is the denominator.

Belkins counts unique replies divided by emails sent, "excluding auto-replies and bounce notifications", across 7,530,489 emails sent through its own managed client campaigns between January and December 2025, producing 34,393 replies and an overall reply rate of 0.45%. Instantly's platform-wide average over roughly the same period is 3.43%, and its report states no exclusion for auto-replies or out-of-office messages anywhere. That is a 7.6 times gap between two numbers with the same name, and we have walked through where that gap comes from before.

Out-of-office messages are the reason it matters here. One reply pool contains them and the other does not, so the two are not measuring the same event. Neither vendor publishes auto-replies by step, so how much of the gap that accounts for is unmeasured, and anyone telling you it explains all of it is guessing.

So the honest reading is not that one vendor is wrong. It is that "where do replies come from" has two answers depending on whether an automatic out-of-office counts as a reply.

Which is the practical reason not to run your sequence on replies at all. You cannot reproduce either vendor's definition in your own inbox, and the version you can count by hand, anything that arrives and looks like a human wrote it, matches neither. Count meetings instead. It is one number, you already know what it means, and the next two sections are about why it points somewhere different from the reply count.

What it costs to earn a reply, step by step

Smartlead publishes the number that makes this concrete, built from "more than 850 million emails sent through Smartlead" in the first half of 2026: how many people you have to contact to earn one reply, by position in the sequence. Email one, 85 contacts. Email seven, 303. Those two endpoints are on its public benchmark page and the steps in between are in the report itself, which counts a genuine human reply against delivered emails and leaves out automatic out-of-office messages.

Its own framing is blunt: "the first email does most of the work". Smartlead also reports that the typical reply lands in about ten minutes, and that follow-ups sent days later are "mostly reaching people who have already decided not to respond."

Belkins' per-step reply rates say the same thing at a different scale: 0.59% at step one, 0.30% at step three, 0.34% at step four, 0.28% at step six. Belkins flags that small rise at step four as real rather than noise, and reads it as evidence that a fourth touch with a genuinely different angle beats a third that reads like a copy of the second.

Every dataset that publishes anything by position agrees on the direction. Each additional email earns fewer replies than the one before it. If replies were the goal, the case for a long sequence would already be closed.

Replies fall. Meetings do not.

This is where the advice on page one stops being useful, because almost none of it separates a reply from a good one.

Saleshandy is the only source found that publishes both counts by step. Across "53.1 million cold emails and 60,000 cold email sequences" sent between January and June 2026, measured against delivered emails rather than sent, the first follow-up produced 116,920 replies of which 15,626 were classified as interested. Saleshandy classifies auto-replies and out-of-office messages as replies and excludes them only from the interested count, so its reply column and Belkins' are not the same measure either. The second produced 60,473 replies and 6,436 interested. The third and beyond, 30,063 and 4,276. Read as a share of every positive reply in the dataset, the first follow-up alone carries 26.41%, and follow-ups together carry 44.5%.

Belkins reports the equivalent in booked meetings, and the shape is sharper: "Step 3 is the single most productive email for booking appointments: it generates more meetings than steps 1 and 2 combined." Over 53% of its email-sourced meetings come from step three or later, and the appointment rate stabilises at 0.2% from step three onward, twice the rate of steps one and two. Belkins' own explanation is a filtering effect. The people still in your sequence at step three are the ones who never opted out and never said no.

Gong measured the same divergence from the copy side, on follow-up language specifically: "We've used phrases like 'I never heard back' in cold email outreach follow-up emails. While this wording increases reply rates, it decreases meetings booked by 14%."

Three separate datasets, three different metrics, one conclusion. Reply volume falls with every follow-up, and so does the absolute number of interested replies: 15,626, then 6,436, then 4,276. What holds up is the rate at which the people still in the sequence convert. Belkins' appointment rate doubles from step three and stays there, and over half of its email-sourced meetings arrive at step three or later. If you are cutting your sequence because reply rates drop after step two, you are cutting the part that books the meetings. The distinction between a reply and a positive reply is the whole decision here.

Where the published numbers come from

The cadence ladders are the weakest part of everything ranking for this question, and it is worth knowing before you build a sequence on one.

Across the five pages currently on page one, forty-two separate timing and count instructions are stated. Two of them reach a measured original and both are traced below. Every instruction about intervals reaches nothing at all. First follow-up after two days, three days, three to four business days, two to four business days, or one week. Total sequence of three to five, four to six, four maximum, five to six, or seven to eight before you are "venturing into annoying territory". Google's AI Overview on this query prints one of those ladders as the answer.

Two of the numbers in circulation do have origins, and both are worth handling carefully.

The often-repeated claim that one extra follow-up lifts replies by 65.8% comes from Backlinko and Pitchbox, who analysed 12 million outreach emails. That study's population is blogger and publisher outreach, guest posts, roundups and link building. It is a real finding about a different job.

The claim that four to seven emails triples your responses traces through Propeller CRM's 2018 post to a Woodpecker study of over 20 million emails, which reported 9% against 27%. Two things about that pair. It is measured per recipient, not per email, so it is not comparable to any reply rate quoted above. And it compares total emails per sequence, one to three against four to seven, not follow-ups. Woodpecker's own current page has dropped the finding.

One more, because it affects a number you may already be carrying: Woodpecker's 2026 statistics page labels a step table as "(Belkins data)" and prints a 3.43% first-email reply rate and a sample of "16.5 million emails". Belkins publishes 0.45% overall, 0.59% at step one, and 7,530,489 emails. The 3.43% is Instantly's platform average, which the same page attributes to Instantly a few paragraphs earlier.

The useful conditional number in this space is older and narrower than any of them. Yesware, from more than 500,000 sales emails: if you get no reply to the first email, you have a 21% chance of a reply to the second, and if that one is also unanswered, a 25% chance of hearing back eventually. The denominator there is the unanswered thread, not the prospect and not the email.

What the rules actually limit, which is not the count

No regulator and no mailbox provider publishes a cap on how many times you may follow up.

In the UK, PECR regulation 22 restricts electronic mail marketing to individuals, and the ICO states that "You can email or text any corporate body (a company, Scottish partnership, limited liability partnership or government body)", while treating sole traders and some partnerships as individuals. It asks for a working opt-out acted on promptly and a suppression list as good practice. It says nothing about message count. The ICO's page also carries a notice that the guidance is under review following the Data (Use and Access) Act.

In the US, CAN-SPAM is an opt-out regime with a primary-purpose test, and each violating email carries a penalty of up to $53,088. We did not read the FTC's full requirements list at source, so no opt-out window in days is stated here.

Google's sender guidance is about consent and complaints rather than counts: don't message people who did not sign up, consider unsubscribing recipients who never open, and keep the spam rate in Postmaster Tools below 0.10%. Nothing we could find from Google, and nothing from any provider, addresses repeated unanswered sends to the same person as such.

The constraint on your sequence length is commercial, not legal. It is what a fifth unanswered email does to the account you wanted to sell to.

A sequence you can defend

Four or five emails, spaced three to four days, is where the first-party sources overlap. Instantly advises at least four and puts its sweet spot at four to seven touchpoints. Belkins puts the practical range at three to five steps. Saleshandy's successful sequences ran four to six emails over 20 to 21 days, and it is the only source publishing intervals by seniority.

Stop at four unless the account is worth a fifth. Belkins reports minimal incremental value at step six, at 0.28%, and Smartlead has the cost of a reply tripling by email seven.

If three steps have produced nothing at all, Belkins' own recommendation is to change channel rather than send the fourth immediately, on the basis that its LinkedIn reply benchmark of over 7% beats continuing down a thread that has gone silent. That is a rule for a dead thread, not an argument against the fourth email on a sequence that is getting opens and replies.

And judge the whole thing on meetings. Every dataset here shows replies falling with each step and outcomes holding or rising. A sequence tuned on reply rate will be too short, and it will be cut in exactly the place where the meetings were.

None of this decides the thing that actually moves your numbers, which is who is on the list and what the first email says. Four or five touches is a range, not a lever. The length and shape of the first email does more work than any interval you pick, and both do less work than sending to the right hundred people.