Email Open Rate: Benchmarks and How to Improve
What open rate means now that a share of opens are machine-generated, and what to use in its place for decisions.
On this page
The short version
- An open is a loaded tracking pixel, not a read. That was always an approximation and is now a mixture of humans and machines in an unknown ratio.
- It is still the best available detector of a sudden delivery problem, because a placement failure shows up here before it shows up anywhere else.
- Published benchmarks for open rate collected before and after privacy protection are not comparable, and most tables do not say which they are.
Open tracking works by embedding a small image and recording the request when a client loads it. That has always meant an open is really "images were loaded", which was a decent proxy while image loading correlated with reading.
It stopped correlating when mailbox providers began fetching images on the recipient's behalf, for privacy. The pixel loads whether or not anyone opened the message, and it loads from the provider's infrastructure rather than the reader's device.
What an open now counts
One of: a person opened the message and their client loaded images; a privacy service pre-fetched the images without anyone reading it; a security scanner fetched them while checking the message. You cannot tell these apart from the outside, and the mix varies by audience.
What it still tells you
Sudden movement. A rate that has been stable for a year and falls sharply is signalling something real — usually a delivery problem, occasionally a broken template or a tracking failure.
That is a genuinely useful property, because deliverability problems otherwise announce themselves slowly. Open rate is the earliest visible symptom most senders have, and it is worth keeping on a chart for that reason alone.
Small drifts are a different matter. A rate declining three points over a year is more likely to be a shift in what devices your list uses than anything you did.
What a drop is usually telling you
Ordered by how often each cause turns out to be the answer when a sender investigates a fall.
Indicative ordering for triage, not measured frequencies. The point is the order to check in: subject lines are the last thing to suspect, not the first.
Benchmarks, and why they disagree
Published open rate benchmarks vary enormously between sources, and the reasons are structural rather than sampling noise. They mix single and double opt-in lists, wildly different list sizes, different definitions of an active subscriber, and — since privacy protection arrived — data collected under two different measurement regimes.
Any table that reports open rate by industry without stating which period the data covers is comparing two incompatible things. The safe use is directional: they tell you roughly what order of magnitude is normal, and nothing more precise than that.
Your own last twelve sends are a better benchmark than any published figure, because they hold constant everything the published figure varies.
Improving it, honestly
The levers that genuinely move opens are, in order of effect: reaching the inbox at all, sending to people who want it, and the subject and preview pair. Most advice concentrates on the third because it is the most fun to work on and the easiest to write about.
Removing unengaged subscribers raises open rate immediately and for two separate reasons — the denominator shrinks and placement improves. It is the single most reliable intervention, and it makes the headline number go down before it goes up, which is why it gets deferred.
Before concluding anything from an open rate
- The series is broken at the point privacy protection changed the measurement
- It has been split by receiving domain
- The share of the list that is Apple Mail is roughly known
- Bot and scanner opens are filtered where the platform allows
- The comparison is against your own history, not a published table
- No content decision rests on it alone
Frequently asked questions
Can I exclude machine opens?
Partially. Some platforms filter opens that arrive within a second of delivery or from known proxy ranges. It reduces the contamination without removing it, because privacy pre-fetching is designed not to be distinguishable.
Is a 20% open rate good?
It depends on list size, consent method, cadence and audience — a small double opt-in list should be far higher, a large single opt-in list lower. The question that resolves is whether yours is higher or lower than it was three months ago.
Does open rate affect deliverability?
Engagement affects placement, and providers measure engagement with signals they can see directly rather than with your pixel. So the underlying behaviour matters; your recorded open rate is a proxy for it, not the thing itself.