You can write the best message of your career and it will do nothing at all from a domain the providers have decided to distrust. Everything else is downstream of that.
What the providers actually require
Most of what circulates about this is folklore. The rules that do exist are published, and they are narrower and stranger than the advice built on top of them.
01
Google, since February 2024
Bulk senders must keep spam complaints below 0.30%, and are told to aim under 0.10%. From November 2025 enforcement moved from temporary deferrals to permanent rejection.
02
Microsoft, since May 2025
Unauthenticated mail at volume is rejected outright rather than filed as junk. Microsoft publishes no numeric complaint threshold at all, so any specific Microsoft percentage you have been quoted was invented.
03
Neither publishes a warmup ramp
Google's entire published guidance is to start low and increase slowly. Every day-by-day schedule in circulation is a vendor's convention. Ours is ours, and we will tell you exactly what it is.
04
Google permits 2,000 a day. We send 30.
The provider's own ceiling is roughly sixty times what we will send from a mailbox. That gap is the difference between an inbox that lasts and one that burns.
Inbox health
- Inbox health is scored continuously. 85 is the floor, we run at 90 to 95.
- Below 85 an inbox sends nothing cold. Volume reroutes to a healthy one rather than being forced through.
- A score dropping more than five points in a day halves that inbox automatically while it is looked at.
- Sends are spaced at random intervals. A perfectly regular sending pattern is itself a signal.
- Placement is tested before every new sequence and after any infrastructure change.
Copy physics
Separate from whether the writing is any good. This is the shape a cold email has to be to get through.
Seventy-five words, plain text
Sentences under twelve words. No HTML, no bold, no bullets, no exclamation marks. Formatting is what bulk mail looks like, and there is a great deal of bulk mail.
No links in the first email
A link in a first touch from an unknown domain is among the cheapest things for a filter to act on. It can wait until there is a conversation.
No two emails share a fingerprint
Consecutive messages are rejected if they share more than five identical consecutive words. Repeated blocks across a domain are trivially detectable.
Tracking off by default
No open pixels, no click wrapping. Both cost deliverability, and open rates have been unreliable since Apple started pre-loading images. We optimise for replies.
When something goes wrong
It will. The thresholds and the actions are set before a campaign starts, so the response happens whether or not anyone is awake.
- 01Complaints between 0.10% and 0.29%Cold volume halves for 72 hours automatically, and copy relevance and list targeting get reviewed before anything scales back.
- 02Complaints at 0.30% or aboveHard pause on the whole domain, not just the mailbox. List purge, segment review, placement tests, and a person has to put it back into rotation.
- 03A Microsoft authentication errorThat mailbox pauses on its own and gets reviewed by hand. No automatic retry, ever. Retrying into a filter that just rejected you turns a warning into a block.
- 04Two attempts, then the domain is retiredAfter a controlled restart at a fifth of previous volume, an identity that fails again does not get a third try. A fresh domain gets provisioned instead.