Two Reports Measured Cold Email the Same Way and Landed 7.6x Apart

Say your last campaign replied at 2%. Whether that is a good quarter or a bad one depends entirely on which report you read next.

Belkins puts the average at 0.45%. Instantly puts it at 3.43%. Hunter puts it at 4.5%. Cleanlist says 3.1%.

The usual explanation is that everyone divides by something different, and that is partly true. It is not the interesting part.

Belkins counts replies against every email it sent. Instantly counts replies against every email it sent. Same denominator, same year, stated in both reports. They are 7.6 times apart.

What is the average cold email reply rate?

Published averages for 2025 and 2026 run from 0.45% to about 4.5%, and the honest answer is a range rather than a number. Cold email reply rate and cold email response rate are the same metric under two names, and the reports use them interchangeably, which is the first small way the numbers drift apart. Belkins reports 0.45% across 7.5 million cold emails sent through 2025, counting replies against every email sent. Instantly reports 3.43% on that same denominator, across platform data from 1 January to 18 December 2025. Hunter reports a 4.5% sequence reply rate across 31 million emails sent by its users in 2025.

One reply per 222 emails, or one per 22. Same metric, same year, same definition.

Why do published cold email benchmarks disagree so much?

Published cold email benchmarks disagree for two reasons: they divide by different things, and they sample different senders. Belkins makes the first point itself, on the same page as its number: a 5% rate against openers and a 0.45% rate against total sends can describe the same campaign. Credit where it is due, and it is only half the answer.

Take the denominator first. Send a hundred emails. Twelve bounce. Of the eighty-eight that land, thirty get opened. One person replies.

That single reply is a 1% reply rate, a 1.1% reply rate, or a 3.3% reply rate, and all three are honest arithmetic. Divide by the hundred you sent, the eighty-eight that arrived, or the thirty that were opened. Nobody is inventing anything. They are picking a denominator, and the worse your deliverability, the more the third number flatters you for it.

The second reason is the one no report volunteers, because volunteering it means conceding that your own number is not the market. It is the sample, and it survives the first. Belkins and Instantly both publish replies over total sends, for the same year, and land 0.45% against 3.43%. No denominator explains that. What explains it is who is in the data.

Belkins is an agency. Its number comes from campaigns it chose to run, for clients who could afford an agency, in the industries that hire one. Instantly is a platform. Its number comes from whoever bought the software and pressed send. Hunter's 31 million come from Hunter users. None of these is a sample of the market. Each is a sample of a customer base, and yours is not in any of them.

Lavender is worth a note. It publishes no single average at all, only rates by department, from 3.2% for finance buyers to 5.2% for technical ones. The "5% cold email reply rate" you see quoted is usually the top of that range with the department removed.

What is a good cold email reply rate by industry?

There is no single good rate, and Belkins' own breakdown shows why. Across the same 7.5 million emails, on the same denominator, food and beverage replies at 3.47% while construction, financial services, healthcare and legal all sit near 0.56%. Belkins puts the spread between its best and worst sectors at almost ten times.

The recipient's company size moves it further, and in one direction.

Same sender, same method, same year. Emailing a ten-person company works three times better than emailing an enterprise.

Neither of those splits explains the Belkins-Instantly gap on its own, and we cannot see inside either dataset to say which mix produced which number. What they do establish is the size of the effect. You do not need a different denominator to move a reply rate threefold. You only need to email different people.

Is a 0.45% reply rate really the honest number?

The 0.45% figure is the most conservative published cold email reply rate that states its own method, measured by Belkins across 7.5 million emails sent through 2025. It is not the lowest number available. It is the one whose denominator you can check.

It is also a number Belkins had to explain. Their reply rates, in their own words, "look dramatically lower than the figures we've published in previous years" because they moved from measuring against people who opened the email to measuring against everyone who received one.

That is a vendor publicly explaining that its own historical benchmarks were inflated by a denominator. Those older numbers are still being quoted.

Belkins published a number that makes its own product look worse, and showed the working. That is rarer than the number.

Why are cold email open rate benchmarks unreliable now?

Cold email open rates are unreliable because the instrument that produces them damages what it measures. Open rates are tracked with an invisible pixel, and Belkins stopped using theirs on the grounds that it "was hurting deliverability throughout the industry in 2024".

Apple had already blunted the other end. Mail Privacy Protection pre-loads images before anyone opens anything, and Hunter notes corporate security filters prefetching mail on top of that, so a share of every open rate since 2021 records a machine rather than a person. An industry spent a decade optimising subject lines against a metric that iOS was partly generating on its own.

Hunter still publishes 30% as the average open rate across its 31 million emails. Treat that as the best available figure from a method that is itself the problem.

For your own campaigns, that makes open rate a vanity metric: it moves every day, and no decision should follow from it. Track replies instead. The same test, applied to the numbers in a pipeline review, is in Sales Metrics That Matter in 2026.

What actually moves a cold email reply rate?

Smaller lists and a third follow-up, not more volume. Hunter's data, across 31 million emails, breaks it down two ways, and both point away from sending more.

Campaigns to between 21 and 50 recipients reply at 6.2%. Campaigns to 500 or more reply at 2.4%. Same platform, same year, and the smaller list outperforms the larger one by more than two and a half times.

Follow-ups move it the other way. One email replies at 3.3%. Three emails reply at 6.8%. Beyond four, the average falls again.

So the two levers with evidence behind them are a shorter list and a third follow-up. The lever with no evidence behind it is a longer list.

Hunter does not say why the smaller list wins. A list of thirty is one a rep can research person by person, and a list of five hundred is not, which is the case we make in our research on whether AI SDRs work.

What does this mean if you plan outbound by volume?

At 0.45% of total sends, 10,000 emails produce roughly 45 replies. At Hunter's 4.5%, the same 10,000 produce 450. Your real number depends on whose customer base you most resemble, which is not something any of these reports can tell you.

Pick the row you are planning against and then ask which column you are entitled to. If the answer is the left one, doubling the sends doubles the replies and costs the same domain reputation twice.

We have published the other half of this argument separately. Bought contact lists decay faster than most teams plan for, and a category filter cannot tell you whether a company is in the market this quarter. That piece is Is Number of Leads a Good Pipeline Metric?. The same test applied to ten other numbers in B2B sales is in the sales statistics fact check.

How should you benchmark your own cold email?

Against yourself, with a denominator you wrote down, because no published average is comparable to your campaign unless you know whose customers produced it.

1. Divide replies by total emails sent. That version cannot be flattered by bad deliverability.
2. Exclude out-of-office and bounce notifications from the numerator.
3. Record the date, the sample size and the segment alongside the rate.
4. Compare this month to last month, not to a vendor blog.

Being straight about the limits of this piece: we did not run these campaigns and we have not audited anyone's data. We read five reports. Three state a denominator, one states none, and one publishes no overall average at all.

FAQ

Check the sample, not just the denominator

Five reports, one metric, and the two that measured it identically are 7.6 times apart.

The denominator is the easy half. Ask it first, because a report that will not tell you what it divided by has told you something. But two reports can agree on the arithmetic and still describe different worlds, and the thing separating them is whose customers they happened to have.

That applies to this piece too. The figure is 0.45%. It is replies over total sends, it is 7.5 million emails, it is 2025, and it is an agency's clients. Quote it with all four or do not quote it.