How Apple Mail Privacy Protection distorts email open rates and what to track instead

The open rate rested on a simple mechanism: an image embedded in each message, which records an open when a mail client loads it.

Apple changed that mechanism for people who read email in its Mail app and turn the feature on. The change affects reports, automations and tests that still treat an open as proof that a person read something.

What Mail Privacy Protection does to a tracking pixel

Apple previewed the feature on June 7, 2021, alongside iOS 15, iPadOS 15, macOS Monterey and watchOS 8. Its announcement described two effects in the Mail app: it stops senders from using invisible pixels to collect information about the user, and it masks the user's IP address so it cannot be used to determine their location.

Email platforms describe the practical result in similar terms. Bird's documentation says Apple Mail can prefetch the tracking image in a message without the recipient reading it, and that the image load still counts as an open. Optimizely's documentation says that when the feature is enabled, Apple caches the content and generates automatic opens, so the platform cannot track real opens from those users.

The consequence is that a raw open count includes machine activity. Bird states that mail privacy features can inflate raw open counts, so they do not precisely measure human engagement.

Which email metrics and features it breaks

Optimizely lists the affected areas in detail. Open rate is inflated by automatic opens and loses much of its meaning for Apple Mail users. Click-to-open rate falls.

Automations are exposed as well. Triggers that check for non-openers misclassify more recipients as non-openers and become unreliable. Subject-line A/B tests that pick a winner on opens are distorted. Segments built by activity are less accurate when they rely only on opens.

Two other effects are less obvious. Location can no longer be identified for Apple Mail users who enable the feature, because their IP address is masked. Dynamic content that depends on the moment of opening has its load timing falsified.

Gmail and Outlook are not affected directly, according to Optimizely. Gmail users who read their mail through Apple Mail are affected if they enable the feature.

How platforms filter automatic opens, and where that stops

Some platforms now separate machine opens from human ones. In Bird, each open event carries an is_prefetched field that flags likely machine opens. The dashboard open rate uses non-prefetched opens, and the raw counts remain available.

Optimizely detects automatic opens and excludes them from report KPIs, which lowers the reported open rate. For Apple Mail users, an open counts only once the recipient clicks.

Filtering does not make opens exact. Bird notes that recipients who block images or use text-only clients can read a message without registering an open at all. The error runs in both directions: some real reads go uncounted, and some machine loads are counted.

Bird advises comparing like with like, such as this month against last month, as long as the audience mix is similar.

Which signals still hold up

Both platforms point to actions the recipient takes. Bird presents clicks, replies and conversions as stronger engagement evidence than image loads. Optimizely determines opens for Apple Mail users from clicks only, and recommends clicks or conversions instead of opens as a criterion.

Clicks are not clean either. Bird cautions that automated link scanners can also generate clicks.

Opens are not discarded entirely. Bird says they remain useful for comparing periods or variants when the audience mix is similar, because the machine noise affects both sides of the comparison. It recommends treating them as a soft signal for trends rather than an absolute or per-recipient measure.

Treat opens as a soft signal: good for spotting trends and large relative differences, unreliable as an absolute measure.

What to change first in an email program

The platform guidance points to a short list of changes.

First, review automations and segments that target non-openers. Bird advises against removing recipients on the basis of opens alone, and suggests combining durable signals with an inactivity window that fits the sending cadence.

Second, move triggers, segmentation and optimization to clicks or conversions instead of opens. Optimizely gives this as its recommendation, and notes that its send-time optimization already uses click behavior for affected recipients.

Third, change how tests are run. Optimizely recommends larger segments for A/B and subject-line tests, excluding detected Apple Mail users from open-based tests, and using the adjusted open rate as the criterion. Bird recommends comparing like with like, such as month over month or variant against variant with similar audiences.

Fourth, estimate how exposed the list is. Optimizely describes one method inside its own product: counting absolute opens grouped by operating system and version to approximate the share of Apple Mail users.

What the sources do not establish

None of the documentation reviewed gives a figure for the share of opens that Apple Mail inflates, and the Optimizely page gives no rollout date or usage percentages.

The material also does not measure which metric best predicts revenue. Bird ranks clicks, replies and conversions above opens as evidence of engagement, but that is guidance on reliability rather than a tested link to sales.

The vendor pages describe their own products. Behavior differs between platforms, and Apple's announcement was a preview of the feature rather than a technical specification.