Industry-standard email open rates: why the benchmark you found is probably wrong

28 August 2026 · 1 min read · The Bulk Mail team

Search for a standard email open rate and you will find confident, specific, mutually contradictory numbers. They are not all lying. They are measuring different things, on different populations, in different years — and most of them predate the change that broke the metric.

Four reasons published benchmarks disagree

1. Privacy features inflate opens

Some mail clients now pre-load remote images on the recipient’s behalf, whether or not a person ever looks at the message. Every one of those registers as an open. Any benchmark that spans this change is comparing two different definitions of the word.

2. Image blocking deflates them

In the other direction, plenty of clients and corporate gateways block remote images by default. Those reads are invisible. Real open rates are therefore both over- and under-counted at once, in proportions that vary by audience.

3. Unique opens or total opens

Some tools count a person once, others count every time the message is displayed. That single definitional choice moves a reported rate substantially, and benchmark tables rarely say which they used.

4. Whose lists?

A benchmark is an average over whoever happened to be using that vendor. Double opt-in newsletters and purchased lists both land in the same average. Your audience is not that average.

What to do instead

• Benchmark against yourself. Your open rate last quarter, measured the same way, is the only comparison where the methodology is held constant.
• Watch direction, not level. A 20% drop in your own rate is a real signal; being 4 points under someone’s published average is noise.
• Check deliverability before copy. A sudden fall is almost always delivery, not subject lines.
• For one-to-one email, drop rates entirely. One person opening one proposal three times is information; a percentage is not.

We deliberately do not publish an open-rate benchmark table. Any number we produced would be an average of our own users, measured our way — which is exactly the problem described above.