What is a machine open?
A machine open is a recorded email open that was caused by software rather than by a person reading the message. Open tracking works by embedding a small invisible image, called a tracking pixel, in the email. When the image is requested, the platform logs an open. Any software that fetches that image produces the same record as a human reader would.
The largest source is Apple Mail Privacy Protection, which prefetches images in every message delivered to a protected user, whether or not the message is ever opened. Corporate security scanners, link checkers and spam filters that render message content produce the same false positive.
How many opens are machine opens
Measurements through 2026 put the machine-generated share somewhere between 58% and 65% of all recorded opens, depending on audience mix and whose panel you read. The exact figure matters less than the consequence: the majority of what an email platform calls an open was not a person.
The share is not uniform across a list, which is the part that does the real damage. It depends on what proportion of recipients use Apple Mail with protection enabled, and that varies by role, by region and by company. Two segments with identical genuine engagement will report different open rates, and the segment with more Apple users will look more engaged.
Why this is worse than a metric being inflated
A number that is uniformly overstated is workable. You adjust the threshold and carry on. Uneven inflation is a different problem, because it corrupts the ranking rather than the level.
Any lead score, trigger or report built on opens therefore ranks contacts partly by their mail client. A sales team working that list top-down is being pointed at Apple users first, and no amount of threshold tuning fixes a ranking that is wrong.
What to use instead of an open
The test for any engagement signal is whether a human had to decide something. A reply, a deliberate click through to a specific page, a meeting booked, a proposal opened more than once, a pricing page visited: each of those costs the recipient effort or attention, which is exactly why software does not produce them by accident.
A click on its own is weaker than it looks, because security scanners follow links too. The distinction worth drawing is between a click on anything and a click through to a page the person had to choose.
Keep opens for one purpose. Aggregate deliverability. If an open rate halves overnight across every segment, something broke: a domain reputation problem, a blocklist, a broken template. The absolute number is meaningless and the sudden change is not. Use it as a smoke alarm, not as a way to decide who to call.
How Empiraa Signal handles machine opens
Signal Score excludes machine opens from the model by design, so a prefetched pixel never moves a score. The model itself is readable: the points each behaviour earns, the cap on any single behaviour, the half-life that decays old engagement and the thresholds where a lead changes band are all visible and editable.
That readability is the point as much as the exclusion. Platform behaviour changes, and a model you can inspect is one you can check again after the next change. A model you cannot read is one where you find out the rankings are wrong by noticing that they stopped making sense.