Cold email benchmarks for 2026: what a good reply rate actually looks like

Two average reply rates were published for the 2025 sending year. Belkins, reviewing 7,530,489 emails sent between January and December 2025, reported 0.45%. Instantly, reviewing billions of interactions between 1 January and 18 December 2025, reported 3.43%. That is a gap of more than seven times, from two sizeable datasets covering almost the same period. Neither figure is wrong.

They disagree because they are not measuring the same thing. Take a benchmark off a blog post, compare it with the number in your sending tool, and you are almost certainly comparing two different fractions and drawing a conclusion from the difference.

The denominator is the whole argument

Belkins counted 34,393 unique replies against 7,530,489 emails sent, with auto-replies and bounces excluded. That is replies divided by individual emails. Instantly measured engaged senders per contact, which is a per-contact figure across a whole sequence.

The distinction matters because one contact receives several emails. A sequence of four to seven touches means the same person is counted once in a per-contact rate and four to seven times in a per-email rate. Dividing by messages rather than by people pushes the percentage down by roughly the number of steps in the sequence, which is most of the distance between 0.45% and 3.43%. You can report either. You cannot report one and compare it with the other.

Before quoting any reply rate, including your own, settle three things:

• The numerator. All replies, unique repliers, or only positive replies? Are out of office messages excluded? "Not interested" is a reply, and it sits inside most published reply rates.
• The denominator. Emails sent, contacts targeted, or contacts successfully delivered to?
• The publisher. Whose dataset is it, and what do they sell?

A reply rate quoted without its denominator is not a benchmark. It is a number with a percentage sign attached.

The headline figures, side by side

Note the last column. Every number in this article comes from a company with something to sell: Belkins is an agency, Instantly is a sending platform, Gong sells revenue intelligence. None of these datasets has been independently audited, none publishes raw data for replication, and each one describes the customers of the business that collected it rather than the market. Treat the figures as order of magnitude anchors, not as a standard.

Two of them are worth holding together. Gong's 344 emails per booked meeting works out at roughly 0.29% of emails sent becoming a meeting. Belkins' 0.45% is a reply rate, from a different dataset and period, so do not divide one by the other and call the result a conversion rate. The useful point is that both land in the same territory: at cold email volumes, outcomes per message are fractions of one per cent, and any plan built on whole numbers is wrong before it starts.

What actually moves the number

Segment size

Belkins found campaigns sent to fewer than 50 recipients replied at 5.8%, against 2.1% for campaigns over 500 recipients. Those are not on the 0.45% per-email basis, so read them as a relative difference rather than a target: nearly three times the reply rate for the smaller segment. This is the largest controllable factor in most programmes, and the one most often surrendered in pursuit of volume. Small segments allow a specific message, and specificity is what earns the reply.

Sequence depth

Belkins reports that steps 2 to 6 produce 58.6% of replies, that step 3 alone produces 35.6% of email-sourced meetings, and that 3 to 5 steps is the sweet spot. Instantly puts the optimum at 4 to 7 touches spaced 3 to 4 days apart, while also reporting that 58% of replies arrive from the first message.

Those last two look contradictory and are not. They are separate datasets with separate definitions, and both point the same way: one email leaves most of the available response on the table, and the gains run out somewhere around the fifth to seventh touch. Write every step as if it will be the only one read, and stop before the follow-ups become their own reason to ignore you.

The ask

The Gong Labs study of 304,174 emails found interest-based calls to action succeeded about 30% of the time, against about 13% for open-ended meeting requests. Asking whether a problem is worth a conversation outperforms asking for a slot in a diary. Gong's wider dataset agrees: pitching in the first email cuts reply rates by up to 57%.

Length

Gong found emails under 100 words and 3 to 4 sentences perform best. Instantly found the best campaigns run under 80 words. Subject lines of 1 to 4 words in lower case performed best in Gong's data. There is no credible dataset in which longer cold emails win.

Open rate is no longer a metric

Open rates are unreliable in 2026. Apple Mail Privacy Protection pre-fetches images and bot clicks inflate the figure further, so a reported open rate mixes human attention with machine activity in unknown proportions. It cannot carry a decision, and it should not appear in a report as evidence of anything.

That has a consequence for one of the figures above. Gong's finding that salesy language cuts open rates by up to 17.9% is measured with the broken instrument, so treat it as directional only. The reply is the first signal in the chain that a person had to generate deliberately.

The deliverability floor caps everything above it

None of these benchmarks are reachable if the mail does not arrive. The Google and Yahoo bulk sender rules took effect in February 2024, and enforcement hardened to SMTP-level rejection in November 2025, meaning non-compliant mail is refused at the door rather than quietly filtered. Microsoft began rejecting non-compliant high-volume mail to consumer domains with error 550 5.7.515 from 5 May 2025. The spam complaint ceiling is 0.3% and the working target is under 0.1%.

Read that as a floor under the whole exercise. A campaign that breaches the sending rules does not produce a worse reply rate, it produces no reply rate, because the denominator never reaches a human inbox. Keep bounces under 2% and complaints under 0.1%, and treat any movement in either as a deliverability problem to fix before touching the copy.

One claim to stop repeating

The widely repeated line that "outbound generates 42% of B2B pipeline" is a corruption of an Instantly statistic about 42% of replies coming from follow-ups. The two statements are unrelated. If the pipeline version appears in a deck, a proposal or your own notes, remove it. The follow-up version is the more useful finding anyway, since it is the mirror image of the 58% that arrive from the first message.

How to set an expectation you can defend

Benchmarks are for sanity checking a forecast, not for making one. Build the expectation from the shape of the actual list.

Belkins' breakdowns are the most useful part of its study here. Reply rates by seniority ran at 0.57% for founders and owners, 0.42% at C-level and 0.32% at VP level. By company size, 0 to 10 employees replied at 0.72%, 11 to 50 at 0.49%, and organisations over 10,000 at 0.22%. By country, the US ran at 0.51% and the UK at 0.48%. Construction, at 0.56% to 0.60%, sat above average. Smaller companies and founders reply more, which is a statement about who reads their own inbox.

Four rules follow:

• Fix the denominator in writing before launch. Agree whether the campaign is judged on replies per email or per contact, and what counts as a reply, at the start rather than in the first review.
• Assume decay. Belkins' average fell from 0.50% in the first half of 2025 to 0.40% in the second half. A benchmark has a shelf life.
• Forecast meetings, not replies. Gong's top 10% of reps book 8 times more meetings than average, so model a new programme against the average rep and not the top decile.
• Recompute before you compare. Rebuild your result on the published denominator first. That single step removes most arguments about whether a campaign is working.

If you would rather hand the operation to someone who runs it this way, that is what Camley Outbound does: lists, infrastructure, copy, sequencing and reporting on a denominator agreed before launch. A four-week pilot is £300, and the ongoing programme is £1,500 a month plus £200 per qualified meeting. Get in touch if you want the method applied to your own market.