Why Open Rates Are No Longer Reliable

For a long time, open rate was the first number anyone looked at. It was easy to measure, it responded to subject-line changes, and it felt like a proxy for attention. That era ended quietly, and a lot of outbound teams are still running tests and making list decisions on a number that no longer means what they think.

What changed

Open tracking has always worked the same way: embed a tiny invisible image hosted on a tracking domain, and record a request for it as an "open". It was never precise — recipients who blocked images were invisible, so the number always undercounted — but the error was at least in one direction and roughly stable.

Apple Mail Privacy Protection inverted that. When enabled, Apple fetches remote content through its own proxy in advance, before and regardless of whether the recipient opens the message. Every tracking pixel in every message loads. The sender records an open for a message nobody read, timestamped to Apple's fetch rather than any human action, from an IP that carries no useful location or device information.

The result is not simply a higher number. It is a number contaminated by an unknown proportion of non-events, where the proportion depends on how many people on that particular list use Apple Mail. That share differs between industries, seniorities and regions, and it differs between two segments of your own list.

Why you cannot just subtract the inflation

The natural instinct is to treat this as a calibration problem: work out the inflation factor and adjust.

It does not work, for a reason worth understanding. The inflation is proportional to the Apple Mail share of the specific recipients who received that specific campaign. Two campaigns to two segments have different shares and therefore different inflation. So does the same campaign sent six months apart, as client usage shifts.

That is precisely the situation in which a metric stops supporting comparison. You can no longer say campaign A outperformed campaign B on opens, because the difference between them may be entirely a difference in who they were sent to. And subject-line testing — the main thing open rate was used for — is exactly a comparison between two groups.

Some platforms offer filtering that flags opens with the fingerprint of proxy pre-fetching. It helps at the margin. It does not restore the metric, because the detection is heuristic, Apple is not the only client that pre-loads, and you are left with a smaller number whose error you still cannot quantify.

The decision that actually gets damaged

The most expensive consequence is not a misreported dashboard. It is list hygiene.

A common and previously sensible practice is to suppress contacts who have not opened anything in several campaigns, on the reasoning that unengaged recipients drag down reputation. Run that rule now and it does something close to the opposite of what you intend: contacts whose client pre-loads images are recorded as engaged whether or not they ever looked, while contacts using a client that blocks images are recorded as unengaged even when they read every message.

You end up pruning the wrong people and retaining machine-generated engagement — a hygiene process that makes list quality worse while appearing to work.

If you suppress on engagement, suppress on replies, clicks and bounces, all of which reflect something that actually happened.

What to measure instead

Reply rate is the metric that survived, and it was always the better one. It requires no tracking pixel, cannot be triggered by a proxy, and corresponds directly to the outcome cold email exists to produce. It is also harder to flatter yourself with, which is part of why it was less popular.

Bounce rate and complaint rate are unaffected and remain the two numbers that decide whether your domain keeps working. Both are reported by receiving infrastructure rather than inferred from an image request.

Click rate still reflects real intent, with a caveat specific to cold outreach: measuring clicks means wrapping links, which replaces your visible destination with a redirect through a tracking domain. That is a recognised phishing pattern and carries a genuine filtering cost on first-touch mail. On a cold campaign, the trade is usually not worth it.

Positive reply rate — replies that are actually interested, separated from the polite declines — is the only number here that correlates with revenue. It requires a human to categorise responses, which is why almost nobody tracks it, and why the teams who do are usually the ones whose outbound works.

What this means in practice

Stop testing subject lines on open rate. If you want to test them, test them against replies, accept that you need more volume for significance, and be honest that most subject-line differences are smaller than the noise.

Stop setting open-rate targets. A 70% open rate is not a triumph and a 25% open rate is not a failure; both are mostly statements about who is on the list.

Keep watching open rate as a trend on a stable list, where a sudden fall still usefully signals that something changed — most often that your mail stopped reaching the inbox. That is a real use, and it is the only one that survives.

And reconsider whether the pixel earns its place at all. On cold outreach it adds a third-party request to a message that is supposed to read like a personal note, in exchange for a number you can no longer act on.

RepMail does not attach open-tracking pixels or wrap links in campaign sends, so a first-touch message carries no third-party requests unless you add one deliberately. Delivery, bounce and complaint telemetry comes from AWS SES itself rather than from an image request, which means the numbers that actually govern your domain's health are measured at the infrastructure layer where they cannot be inflated by a proxy.