# Why Open Rates Stopped Being Reliable

> Image proxying, prefetching and privacy features broke open rates. Here is what an open now measures, what to track instead, and how to compare campaigns.

Canonical: https://www.cresca.xyz/blog/why-open-rates-stopped-being-reliable
Author: Lalit Kumar Jangid
Published: 2026-09-26
Last updated: 2026-10-30

For two decades, open rates were the headline metric of email marketing. They were simple, they were immediate, and everybody had a benchmark for them. They are now unreliable enough that treating them as a primary decision metric will lead you to confidently wrong conclusions. This is not a measurement bug that will be fixed; it is a structural consequence of how modern mailbox clients handle images.

## How are email opens counted?

The mechanism was always indirect. An email contains a tiny invisible image hosted on the sender's server. When the client loads the image, the server records a hit and attributes it to that recipient. It never measured reading. It measured image loading, which was usually a decent proxy for opening, back when clients loaded images on open.

## Why are open rates no longer reliable?

Open rates are no longer reliable because three developments broke the link between image loading and reading: image proxying, prefetching by clients and security gateways, and privacy features such as Mail Privacy Protection. As a result, open rates are inflated, compressed toward a narrow band, and no longer comparable between audiences or over time.

Three separate developments broke that proxy.

**Image proxying.** Clients now fetch images through their own servers and cache them. The recipient's device may never contact your server at all, and a single cached fetch can generate one record for many recipients, or many records for one.

**Prefetching.** Some clients and security gateways load images before the recipient sees the message, sometimes automatically as part of scanning for malware. That registers an open for a message that may be deleted unread.

**Privacy features.** Mail Privacy Protection and similar features load remote content on the recipient's behalf. For recipients who have them enabled, opens are effectively recorded at delivery time rather than when anyone reads the message.

The net effect: open rates are inflated, compressed toward a narrow band, and no longer comparable between audiences with different client mixes or over time as client defaults change.

## What the inflation does to your decisions

Inflated opens cause specific, predictable errors.

- **Subject line testing breaks.** If a large share of opens are recorded at delivery, subject lines cannot differentially affect them. You will pick winners from noise.
- **Segment comparisons mislead.** A segment using clients that prefetch shows a higher open rate than one that does not, regardless of actual engagement.
- **Deliverability signals get muddied.** Engaged-recipient signals are what mailbox providers weight, and an inflated open count overstates how much genuine engagement you have.
- **Trend lines become meaningless.** A rise in open rate after a client update is not a content win.

## Which clients inflate most

Inflation is not evenly distributed across your list, which is what makes comparisons unsafe rather than merely noisy. With Apple's Mail Privacy Protection enabled, remote content is loaded in the background, so messages can register as opened whether or not they are read. Corporate gateways that scan links and images register opens on messages that may never reach an inbox. Webmail clients frequently prefetch for their own caching.

Meanwhile a recipient on a desktop client with images blocked may read a message fully and register nothing. So the segment with the most engaged users can plausibly show the lowest open rate, purely because of client configuration. Any conclusion drawn from comparing two such segments is measuring infrastructure, not interest.

## How can you tell if your open rates are inflated?

Segment by receiving domain and compare open rates across providers. A provider whose open rate is dramatically higher than the rest and clustered close to delivery rate is prefetching for much of your list. You can also compare opens in the first minute after send with opens over the following day.

If your sending tool reports opens, there is a quick sanity test. Segment by receiving domain and compare open rates across providers. If one provider shows open rates dramatically higher than the rest and clustered suspiciously close to delivery rate, that provider is prefetching for a large share of your list. The size of the gap is a reasonable estimate of how much of your open data is fabricated.

A second test is to compare opens recorded in the first minute after send against opens recorded over the following day. Historically the pattern spread across hours. If a large spike appears within seconds of delivery, those are not readers.

## What should you measure instead of open rates?

Measure downstream outcomes such as trial starts, purchases or activated accounts first, then click-through rate, which is far harder to fabricate because it needs a deliberate action. Watch complaint and unsubscribe rates per campaign as guardrails, and treat opens only as a rough directional signal you never make decisions on.

Click-through rate is the closest available substitute and is far harder to fabricate, since it requires a deliberate action. Use it as your primary engagement metric.

Then measure the thing you actually care about. For most teams that is a downstream outcome attributed to the campaign, such as trial starts, purchases or activated accounts. This is the only metric that cannot be gamed by client behaviour, because it measures what the recipient did rather than what their mail client did.

Also watch **complaint rate** and **unsubscribe rate** per campaign. Rising complaints are the earliest reliable warning of a deliverability problem, and they are unaffected by image handling.

A practical hierarchy: conversion first, clicks second, complaints and unsubscribes as guardrails, opens as a rough directional signal you never make decisions on.

## Comparing campaigns honestly

If you must use open rate for continuity with historical reporting, compare like with like. Segment by mailbox provider when comparing periods, because client behaviour drives the number. Compare against your own prior sends to the same segment rather than against industry averages, which are drawn from unknown mixes and are not actionable for an individual account.

Where possible, run holdout groups. A small percentage of a segment that receives no message tells you what would have happened anyway, which is the only way to attribute an outcome to email rather than to seasonality or a concurrent launch.

## How Cresca reports this

Each campaign report shows recipients, deliveries, bounces, complaints and unsubscribes: the figures that image proxying and prefetching cannot distort. Cresca offers open and click tracking as an opt-in setting on each sending domain, counting unique human opens and clicks and leaving out bots. Treat those opens as a rough signal for the reasons above. For engagement, measure the downstream outcome in your own systems, such as sign-ups, purchases or activated accounts, and record it against contacts with tags or custom fields if you want to segment on it.

Because campaigns are queued and delivered through a sending queue, delivery events reflect actual handoff to the receiving server rather than an optimistic send, which makes delivery rate a meaningful reliability metric rather than a count of attempts.

## Pricing

Cresca starts with a Free plan that runs for 30 days with no credit card required, and paid plans raise the limits as you grow; see [current plans and limits](https://www.cresca.xyz/pricing).

## The short version

Open rate measured image loads, and image loading is no longer a proxy for reading. Treat opens as directional at best and never as the basis for a subject line test or a segment comparison. Decide on downstream outcomes, guard with complaints and unsubscribes, and compare only against your own comparable sends.
