Open rate is dead in 2027: a post-mortem of email marketing's favourite metric

Every marketing team has a metric they quote without thinking. For email, it has been the open rate since roughly 2003 — a comforting percentage that proved someone, somewhere, cared. The problem is that the number stopped representing reality on 20 September 2021, and most teams are still quoting it in 2027 as though nothing happened.

This article is not a rant. It is a forensic walkthrough. We will look at the exact mechanism that broke open rate, the cascade of changes at other mailbox providers that made it worse, and the narrow set of circumstances under which the metric is still faintly useful. Then we will talk about what to put on the dashboard instead.

September 2021: the day the metric died

Apple Mail Privacy Protection (MPP) did one simple thing: it routed every image request through a proxy, and it fetched the image even if the user never opened the message. The feature was opt-in by prompt, but the prompt was worded in a way that roughly 95% of users accepted within the first six months.

The instant that shipped, every email sent to an Apple Mail user registered an open — deterministically, without human involvement. Apple Mail had roughly 40% of the email client market at the time. Overnight, a metric that used to land somewhere between 15% and 35% for a healthy list started landing between 55% and 90%.

Marketing teams did one of three things. The disciplined ones segmented out Apple Mail opens and kept tracking the rest. The hopeful ones decided the new higher number was a sign their content had improved. The cynical ones quietly rewrote their KPI targets upward and never mentioned it again. All three approaches are still represented in every ESP dashboard you will touch today.

What MPP actually does under the hood

Understanding the mechanics matters, because it makes clear why the metric cannot recover. When an Apple Mail client receives a message, the following sequence happens before a human is involved:

1. The client downloads the message headers and body to device storage.
2. A proxy service, operated by Apple, fetches every remote resource in the HTML — images, tracking pixels, hosted CSS.
3. The proxy routes the fetch through an IP that does not geographically match the user, strips the User-Agent, and serves the content to the client from its cache.
4. Your tracking server sees an HTTP 200 request for the pixel, records an open, and increments the counter.

Note what is missing from this sequence: the user. The user may never see the message. The user's device may be off. The proxy will still fetch. Every email in every Apple Mail inbox generates exactly one pixel load, once, regardless of whether it is ever read.

The cascade: Gmail, Outlook, and the AI summarisers

If Apple were the only problem, you could segment around it. The trouble is that every other major mailbox provider has since made changes that further pollute the signal.

Gmail

Google has proxied images through Google's own CDN since 2013, but until recently it only fetched on actual user view. From 2023 onward, Gmail began pre-fetching images on the mobile app when messages arrived, to enable offline reading. Pre-fetch rates now exceed 60% of delivered mobile messages according to the public research available.

Gmail also runs its own spam scanning infrastructure that renders the HTML of inbound messages. Each render triggers pixel loads. If your email is flagged for extended scanning, the pixel may fire two or three times before the human ever sees it.

Outlook and Microsoft 365

Outlook has the opposite problem. The default configuration in many enterprise tenants blocks remote images entirely, requiring the user to click "Download images" before the pixel loads. Recipients in those tenants who actually read and engage with your message will not register as opens.

So Apple inflates your open rate by firing pixels on messages nobody reads, while Outlook deflates it by suppressing pixels on messages people do read. The net number you see is a meaningless weighted average of two opposing distortions.

AI summarisers

Since 2024, a growing fraction of mail is read first by an AI agent — Apple Intelligence, Gmail's inbox summary, Microsoft Copilot, third-party assistants. These agents load the full HTML to generate their summary, firing pixels in the process. Whether that counts as an "open" is a philosophical question with no good answer.

When open rate still tells you something

It would be dishonest to claim the metric is completely useless. There are three narrow cases where it retains some signal value:

• Very large relative swings on identical lists. If you send two broadcasts to the same list 24 hours apart and one opens at 45% while the other opens at 15%, something real happened — likely a blocklist event or a bulk folder drop.
• Trend lines on a stable segment. If Apple Mail opens stay constant at 75% while your overall opens slide from 50% to 20%, the slide is real and is happening at Gmail or Outlook.
• Unique opens on very small, curated lists. On a list of 30 VIP investors, you can usually tell who actually read the message. The signal-to-noise is salvageable.

Notice that none of these uses resemble how the metric is typically quoted in a weekly report. "Our open rate was 42% this week, up 3 points" is a sentence that conveys exactly no information in 2027.

The replacement stack

If you retire open rate, what goes in its place? A short, defensible list:

1. Inbox placement rate. Of the messages you sent, what percentage landed in the primary inbox versus promotions, updates, or spam? This is measurable with seed testing.
2. Reply rate (for outbound and cold). Humans reply. Pixels do not.
3. Click-through rate on a tracked link that is not also a tracking pixel. Note the caveat — Apple pre-fetches some links now, too. Use link decorators that require a real redirect.
4. Conversion or revenue per thousand sent. The bottom-line number. Harder to fake.
5. Unsubscribe and complaint rate. Negative signal, but unmistakably human.

How to retire the metric without a fight

Executives like the old number. It was large, it was familiar, and it made people feel good. Replacing it requires care.

1. Start reporting inbox placement alongside open rate for three months. Do not remove open rate yet.
2. In month two, add a footnote to every open rate figure: "approx. 60% of this number is machine pre-fetch."
3. In month three, replace the headline KPI with inbox placement and reply rate. Keep open rate available on request, like a legacy report.
4. When asked why open rate fell, explain that it did not fall — you finally stopped counting robots.

FAQ