Apple has had Mail Privacy Protection since September 2021, in iOS 15, iPadOS 15 and macOS Monterey. Anyone who switches it on, and that is the majority, has Apple fetch the images in a message up front through its own proxy. Including the tracking pixel. The result: an open is registered while the recipient may never have seen the message, and the IP address you get back is Apple's.
What that means for your numbers
- Open rate is inflated, and by how much you cannot tell exactly, because it depends on how much of your list uses Apple Mail
- A/B testing subject lines on opens mostly measures the share of Apple users per variant
- Optimising send time on open timestamps steers on the moment Apple fetches the pixel
- A reactivation segment based on non-openers mails people who do read, and leaves out people who were only opened by the proxy
What I do steer on
Clicks, conversion and unsubscribes. Those are human actions: someone clicks themselves, buys themselves and unsubscribes themselves. They are scarcer than opens, so you need more volume or more patience before you can say anything, but they do point the right way.
In practice that means: click-through rate per campaign, conversion on the landing page with its own UTM per campaign, and the unsubscribe rate as a counterweight. That last one is the most honest signal that you are sending too often or too broadly, and it is rarely reported because it does not look good.
So what about open rate?
I do not remove it from the report, but it sits there as context with a line through it and a short explanation. Throwing it out creates noise when comparing with older periods, and without an explanation someone will still take a decision on it. What does work is reading it relatively: an open rate that suddenly halves within the same list still says something about delivery, even if the absolute number is wrong.
Testing can simply continue
Testing subject lines stays useful, just not on opens. Test on the click and on the conversion after it, and expect the difference to be smaller than you were used to. A difference of a few tenths of a percent on a list of a few thousand people is not a winner but noise. The email marketing demo on this site shows an example of exactly that: a test the data cannot decide, with the human choice and the reasoning next to it.