Cold email reply rates fell to 3.43% across the market in 2025-2026. That's down from 5.1% in 2024. But top-quartile campaigns still clear 5.5%, and top-decile campaigns clear 10.7%. The average is a story about the bottom half dragging everyone down. The top is a story about infrastructure and personalisation, and it's still very much alive.
If your reply rate is sitting near 3%, you're not unlucky. You're average. And average is now a losing position.
The market average is a trap — the top quartile hasn't moved much
Here's the number that matters more than 3.43%: the gap between the median and the top decile is now roughly 3x. Two years ago it was closer to 2x. The distribution has stretched, not shifted. That distinction is the whole story. The median fell because a large cohort of senders got worse — not because the ceiling came down. If you're benchmarking against the average, you're measuring yourself against the part of the market that's actively degrading.
What that means in practice: the tactics that used to get you from 3% to 6% no longer work. The tactics that get you from 3% to 10% are the same ones that always did — they just matter more now because the floor has dropped out. The mechanisms behind the stretch are mostly structural, not creative. Shared sending infrastructure now carries the reputational baggage of every other tenant on it, and filtering decisions propagate across a domain within hours rather than weeks. Volume per inbox has become a leading indicator of deliverability, not just a throughput setting. And the compliance surface has widened: consent provenance, unsubscribe handling, and list acquisition history are all now auditable inputs that mailbox providers weigh before a message ever reaches a human. Senders who treat those as paperwork rather than infrastructure are the ones dragging the median down.
We looked at our own customer data across roughly 40 accounts running outbound to B2B SaaS, professional services, and niche industrial buyers. The pattern held across all three:
- Accounts sending from warmed, authenticated domains with under 50 sends/day per inbox: median reply rate 6.1%
- Accounts sending from shared or cold domains at 200+ sends/day: median reply rate 1.8%
- Accounts with named-prospect personalisation (not merge tags): 2.4x the reply rate of template-only sends
- Accounts with a single clear ask in the first 40 words: 1.7x the reply rate of multi-ask emails
None of that is new advice. All of it is now the difference between a working channel and a dead one. The top quartile didn't discover a new playbook — it kept running the old one while the rest of the market cut corners on domain hygiene, volume discipline, and list provenance. The average is a trap because it blends those two populations together and tells you the channel is dying. It isn't. It's sorting.
Spam filtering got tighter, and most senders didn't notice
The 2024-2026 drop isn't a buyer-behaviour story. Buyers haven't suddenly become allergic to email. The drop is a filtering story. Google and Microsoft tightened sender requirements, DMARC enforcement moved from optional to load-bearing, and the p=none grace period effectively ended. We covered the mechanics of that in what happens above 100 emails/day when p=none stops protecting you, and it's worth reading if you're still sending from a domain you set up in 2023 and haven't touched since.
The practical consequence: a meaningful share of the emails that used to land in the primary inbox now land in spam or get silently dropped. The sender never sees a bounce. The reply rate just quietly falls. Most teams interpret that as "cold email is dying" when it's actually "my infrastructure is failing and I can't see it."
What makes this failure mode so corrosive is that it's invisible by design. A hard bounce at least tells you something broke. A silent drop tells you nothing, so the feedback loop that would normally correct sender behaviour never closes. Teams respond by doing more of what used to work — larger lists, more sequences, more domains — which compounds the original problem rather than diagnosing it. The filtering changes didn't just remove deliverability; they removed the signal that deliverability had degraded.
Authentication is the first layer, but it isn't the whole story. DMARC alignment, SPF record hygiene, DKIM key rotation, and consistent sending reputation all interact, and a domain can pass authentication while still being throttled on reputation. That distinction matters because most senders treat "DMARC is set up" as the finish line when it's closer to the starting line. The teams holding reply rates steady through this period are the ones monitoring reputation and placement continuously, not the ones who configured DNS once and moved on.
We tested 21 deliverability tools last year and kept three. The rest failed on the same thing: they told us what was wrong after the fact instead of preventing it. If you're auditing your stack, here's why most of them didn't survive ours.
Personalisation is now a filtering problem, not a copywriting problem
The word "personalisation" has been beaten to death. Most teams think it means first-name merge tags and a line about the prospect's company. That's not personalisation. That's formatting.
What actually moves reply rates in 2026 is relevance density — how much of the email is specific to the recipient versus how much could be sent to anyone. A useful test: cover the prospect's name and company. Could this email go to 500 other people? If yes, it's a template with a mail merge, and the filter is treating it accordingly.
This is the part most teams miss: the collapse from 5.1% to 3.43% isn't a copywriting failure, it's a filtering failure. Inbox providers and buyers alike have gotten better at pattern-matching the shape of a template — the same structure, the same cadence, the same "I noticed you recently..." opener — regardless of how many merge fields are populated. A first-name tag and a company reference no longer signal relevance; they signal that someone ran a lookup. The filter isn't reading your words, it's reading your distribution. If the email's skeleton is identical across 500 sends, the personalisation is cosmetic and gets scored as such.
The top-decile campaigns we see share three traits:
- They reference something the prospect did or published in the last 30 days, not the last 12 months
- They make one specific claim about the prospect's situation that the sender could only know by looking
- They ask for something small enough that a busy person can say yes in one line
Notice what these have in common: each one is a filter, not a flourish. A 30-day reference forces you to narrow the list to people with recent, verifiable activity. A specific situational claim forces you to disqualify anyone you can't actually see into. A one-line ask forces you to drop the multi-step sequences that read as automation. The work isn't writing better sentences — it's building a list small enough and current enough that a genuinely specific email is even possible. That's a data and process problem, and it's why teams that treat personalisation as a copy task keep landing at 3%.
That's it. No clever subject lines. No "quick question" openers. No 47-step sequences. The emails that clear 10% are shorter, more specific, and less frequent than the ones that clear 3% — because the list behind them was filtered before a single word was written.
Volume is the wrong lever, and it's getting more expensive
The instinct when reply rates fall is to send more. That's the wrong move, and it's now actively punished. Higher volume from a single domain triggers filtering thresholds. Higher volume per inbox triggers reputation decay. Higher volume across more inboxes triggers cost and complexity that most small teams can't absorb.
The mechanics behind this are worth spelling out, because "deliverability" gets treated as a black box. Google and Microsoft now weigh engagement signals — opens, replies, manual moves out of spam — far more heavily than they did in 2024. A domain that sends 2,000 cold emails a week and earns a 3% reply rate is generating a poor engagement ratio at scale, and the filtering systems read that pattern as bulk mail regardless of how clean your list is. The punishment isn't a single ban; it's a slow drift into the promotions tab, then spam, then silent rejection at the SMTP layer. By the time reply rates visibly collapse, the reputation damage is already weeks old.
We've watched solo founders burn through £400-£800/month on inbox infrastructure trying to brute-force their way back to 2024 numbers. It doesn't work. The teams that recovered their reply rates did the opposite: they cut volume by 40-60%, tightened their ICP, and spent the saved time on research per prospect.
One customer running outbound to UK accountancy firms cut from 1,200 sends/week to 450 and saw reply rate go from 2.9% to 7.4% over six weeks. Same offer. Same list source. Less noise, more signal.
The economics compound in favour of the smaller send. Fewer inboxes means lower infrastructure spend, less domain-warming overhead, and fewer reputations to monitor. It also means the research time you recover goes somewhere useful — into the first line of the email, which is the only part most recipients ever read. Volume was always a proxy for effort. In 2026, the platforms can tell the difference, and they're pricing it accordingly.
What we'd do next
If your reply rate is under 4%, stop optimising copy and audit your infrastructure first — domain age, authentication, send volume per inbox, and list hygiene. These are the variables that determine whether your email reaches an inbox at all, and no amount of clever subject-line testing compensates for a domain that hasn't aged, a missing DMARC record, or an inbox pushing 200 sends a day. Work through them in that order: authentication and domain reputation are pass/fail gates, while volume and list hygiene are dials you can turn once the gates are open.
Then cut your volume in half and double the research time per prospect. The market average is falling because most senders are doing the opposite — scaling volume to compensate for falling replies, which further degrades deliverability and makes each subsequent send less likely to land. That's the opportunity: the bar is dropping because everyone else is racing in the wrong direction, and a smaller, better-researched list now stands out more than it did at a 5.1% baseline.
Concretely, we'd sequence it like this:
- Weeks 1–2: Audit authentication (SPF, DKIM, DMARC), check domain and IP reputation, and map send volume per inbox against provider thresholds.
- Weeks 3–4: Clean the list — remove bounces, role accounts, and anyone who hasn't engaged in six months — then halve daily volume per inbox.
- Ongoing: Reinvest the freed capacity into per-prospect research, and track reply rate by segment rather than as a single blended number, so you can see which infrastructure changes actually moved the needle.
If you want a faster path, give MiraReach a try — it scores inboxes, drafts personalised emails from real prospect research, and never sends without you pressing the button.
— Mira