The open rate said 28%. Gmail said 11%.

Two numbers, one account

Since 2021, Apple Mail Privacy Protection pre-loads message content for its users, and every pre-load registers as an open. Klaviyo counts those the same as any other open, because on the wire they are the same. On this account Apple is 16% of delivered mail and reads a 67% open rate — machines opening on schedule, not people reading. In the last two weeks Apple accounted for 38% of every open on the account.

Gmail does not auto-open. Its opens are humans, and on this account they read 18% against Apple’s 67%. The account-wide open rate blends both — plus nineteen smaller providers — into one figure near 32% that describes nobody.

The fall the average hid

Over three weeks, Gmail’s human opens fell from 18% to 11% — a drop of more than a third, at the provider carrying 44% of this account’s mail. The account-wide open rate slipped from about 32% to about 28%.

The arithmetic is why. Gmail’s seven-point fall enters the blend at 44% weight — about three of the four points the average lost — and Apple’s machine opens do not fall with it, so the average barely registers what the biggest provider on the account is doing. A dashboard watching the blended rate sees noise. The list is telling Gmail something and the average is repeating Apple.

What to read instead

Per-provider opens, each measured with the unique measurement and each judged only against its own trailing baseline. Never against each other: Apple’s auto-opens and Gmail’s image proxy make cross-provider comparison meaningless by construction, which is why SpamSpike refuses to rank providers.

On this account the provider panel prints one sentence when this happens: opens at Gmail fell, over how many days, on how much delivered mail — human opens, not the Apple-inflated account rate. That sentence fires only for providers whose opens are human; the machine-open set is gated out of it structurally.

What we would not claim

A falling Gmail open rate co-occurs with list fatigue, with tab reshuffles, and with heavier templates — Gmail clips messages over about 102KB, and a clipped message hides the open pixel entirely. The drop is a reason to look, not a verdict on why. We do not print “your list is rotting” from an open curve, for the same reason the incident pages do not name a send without a feedback-loop identifier: a confident wrong cause is worse than an honest open question.

What the number does earn is attention. An account that watches only its blended open rate has no instrument that can see this happen — the instrument averages the signal away.