What actually changed
If your open and reply rates fell off a cliff over the last year and no amount of subject line testing brought them back, you are not doing it wrong in the way you think. The ground moved. Here is what moved, and what still works on the new ground.
Cause one: deliverability physics tightened
Google and Microsoft spent the last two years raising the bar. SPF, DKIM and DMARC went from nice to have to mandatory. Sender reputation is scored more aggressively, and new domains earn trust more slowly. An email that would have reached the inbox in 2023 now needs correct authentication and a warmed domain just to clear the gate.
Most teams that see rates collapse are sending from domains that were never set up for it, often the main company domain, with no warmup and no monitoring. That is not a copy problem. It is an infrastructure problem, and it has a known fix, which we wrote up in full in the deliverability playbook.
Cause two: everyone's email looks the same
When one AI model writes a thousand personalised emails, the output converges. The same structures, the same openers, the same rhythm. Content filters cluster on exactly that kind of statistical sameness, so mass AI personalisation does not beat the filters, it feeds them. The reply data backs this up: AI written cold email closes most of the copy gap to humans but loses on deliverability faster than it loses on writing.
The uncomfortable part is that sending more, which was the old lever, now makes this worse. Volume plus sameness is the exact signature spam systems are trained to catch.
Cause three: your list went stale
B2B contact data decays at roughly a quarter to a third per year as people change jobs and companies pivot. A list bought or exported once and worked for months is mostly wrong by the end, and wrong addresses bounce. Bounces are the loudest negative signal a mailbox provider reads, so a stale list actively damages the reputation that gets your good emails delivered.
What works on the new ground
The teams still getting 15 to 25 percent reply rates in 2026 did not find better copy. They changed the model. Instead of sending a lot to a static list, they send a little to accounts that are in motion right now: a company that just raised, just hired a first sales rep, just visited a pricing page. Reaching a buyer during a real moment, with correct infrastructure behind the send, is what the volume era was crudely approximating and mostly missing.
Concretely, that means three layers working together. A signal engine that decides who and when. A sending foundation that actually reaches the inbox. And a human approving what goes out, because the fully autonomous version of this churned 50 to 70 percent of its buyers within a year.
The numbers, plainly
| Approach | Typical reply rate |
|---|---|
| Volume cold email, static list | under 2 percent |
| Signal based, single signal | 8 to 15 percent |
| Signal based, stacked signals | 15 to 25 percent |
The gap between the top and bottom row is not talent or budget. It is whether the email arrives when the buyer is in motion and whether it reaches the inbox at all.
Questions people also ask
Is cold email dead in 2026?
No. Volume cold email to static lists is effectively dead at under 2 percent reply. Signal based outbound with proper deliverability still reaches 15 to 25 percent. The channel works, the old method does not.
Why did my open rates suddenly drop?
Almost always deliverability. Tightened provider rules plus sending from an unwarmed or main domain pushes mail to spam, where it is never opened. Fix authentication and warmup first.
Does AI personalisation help or hurt?
Used for research and targeting it helps. Used to mass produce a thousand similar emails it hurts, because filters cluster on the sameness. The edge is in who you reach, not in generating more copy.