A reply, a click to a specific page, a meeting booked or a proposal opened twice. Anything that costs the recipient a decision, because that is the thing a mail server cannot do by accident.
Apple Mail Privacy Protection prefetches the tracking pixel in messages delivered to protected users, whether or not anybody opens them. Measurements through 2026 put the machine-generated share of recorded opens somewhere between 58% and 65%, depending on the audience and whose panel you read.
So the majority of what your email platform reports as an open was a server. Any score, trigger or report built on that number is built on the behaviour of software.
Why this is worse than a wrong number
A metric that is uniformly inflated is annoying but workable: you adjust the threshold and move on.
This one is not uniform. The inflation depends on what share of your list uses Apple Mail with protection on, which varies by role, by region and by company. Two segments with identical real engagement will report different open rates, and the one with more Apple users will look more engaged. Then the score sends your reps at them first.
That is the actual damage. Not that opens are overstated, but that they are overstated unevenly, so the ranking is wrong rather than just the level.
What to trigger on instead
The test for any signal: did a human have to decide something?
| Signal | Human decision required | Use it |
|---|---|---|
| Open | None. Prefetched | No, except aggregate deliverability |
| Click, generic footer or unsubscribe area | Little, scanners click too | No |
| Click to a specific page, deliberate | Yes | Yes |
| Reply | Yes, clearly | Strongest single signal |
| Meeting booked | Yes, with calendar cost | Strongest intent signal |
| Proposal opened more than once | Yes, and probably circulated | Very strong late signal |
| Pricing page visited, identified company | Yes | Strong, see below |
| Forwarded, inferred from a new recipient | Yes | Strong, rarely captured |
Security scanners click links too, which is why "clicked" on its own is weaker than it looks. The distinction worth drawing is between a click on anything and a click through to a page the person had to choose.
The signal that mostly replaces the open is the one that happens off email entirely: the company turning up on your pricing page. That is covered in a company read three pricing pages and nobody called them, and it is a better trigger precisely because nothing prefetches a pricing page.
What to do about the scores you already have
Most lead scoring models were tuned before this and quietly broke.
Audit what is actually driving the score. Take your top fifty scored leads and work out what earned them the points. If opens are the main contributor, your scoring is a ranking of Apple users.
Re-weight to decisions. Reply, meeting, deliberate click, repeat visit. Opens to zero.
Add a half-life. Engagement from six weeks ago should not count the same as engagement from Tuesday. Without decay, a score becomes a lifetime total and the highest scores end up belonging to your oldest contacts rather than your most interested ones.
Cap any single behaviour. One person opening the same email forty times, which happens, should not outrank a different person who replied.
This is why Signal Score is built the way it is: machine opens never count, and the points, caps, half-lives and band thresholds are all visible and editable rather than inferred by a model you cannot inspect. The argument for that design is the subject of can you read your own lead scoring model, and this is the clearest example of why it matters. If the model is readable, you can check it still excludes the right things after the next platform change. If it is not, you find out when the rankings stop making sense.
Keep one use for opens
Aggregate deliverability. If your 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. Do not use it to decide who to call.
Related reading: can you read your own lead scoring model, and a company read three pricing pages and nobody called them.
Machine open has a definition on this site. See how Signal Score is built, or what Signal costs.
