Insights · Measurement

Opens are lying. Why we split on clicks.

Since Apple Mail Privacy Protection, an “open” often means a server downloaded an image. Here is what that does to your journeys, and what we measure instead.

  • Measurement
  • 5 min read

Email opens are tracked with a tiny invisible image. When the image loads, Marketing Cloud records an open. For years that was a fair proxy for “a person looked at this email”. It no longer is.

What changed

In September 2021, with iOS 15, Apple introduced Mail Privacy Protection. When a user turns it on, Apple Mail downloads the images of every message on delivery, whether or not the person ever reads it. The tracking image loads, and Marketing Cloud records an open that no human made.

At the same time, corporate security filters started scanning incoming email: they open messages and follow links to check them, which creates opens and clicks within seconds of delivery.

Illustrative timeline of opens after a send Machine opens spike in the first minutes after delivery; human opens spread over the following hours and days. Send+2 h+12 h+24 h+48 h Machine opens: images fetched on delivery (Apple MPP) and security scanners, within minutes Human opens: spread over hours and days Machine opens: minutes after delivery Human opens: hours, days Illustrative shape, not real data
Why open-based splits misfire: most machine activity lands in the first minutes, before any human has seen the message. Illustrative shape based on the mechanisms described above.

What it breaks

  • Engagement splits: “opened Welcome #1” sends people down the engaged path even if they never read it.
  • Re-engagement logic: inactive contacts look active, so they never get the win-back message.
  • Reporting: open rates rise while revenue does not, and decisions get made on the wrong number.

What we do instead

Split on clicks, after a wait

In our journeys, decision splits read clicks, not opens, and only after a wait of at least a day, so that the real behaviour has time to happen and the scanner noise is already in the past.

Filter the obvious bots

Clicks that arrive within seconds of the send, or on every link of a message at once, are typical of security scanners. A SQL query on the click data views flags them so they can be excluded from scoring and reports.

Measure what only a person can do

The numbers we put in front of stakeholders are clicks, conversions, replies, form submissions and revenue. Opens stay in the report as a deliverability signal, never as the success metric.

A quick sanity check

If a send shows a very high open rate with most opens in the first minutes, you are looking at machines, not at your audience.

Key takeaways

  1. Treat open rates as directional only.
  2. Make journey decisions on clicks, after a wait.
  3. Exclude scanner-like clicks from scoring.
  4. Report conversions and revenue, not opens.
  1. Sources: University of Colorado, Artificial email engagement rates (Marketing Cloud)

Keep reading

MethodOne signal path, from your data to every channelAIAI in Marketing Cloud, without the hype

Want this inside your instance? Let’s build it together.

Tell us where you are and we’ll tell you what it takes to get there.

Talk to us
Nmc Bubble