Open rate is the percentage of delivered emails that a sending platform records as opened. It is not a count of people who read something. It counts the times a remote image embedded in a message was fetched from a server, and the gap between those two descriptions has widened every year since 2013.

The metric exists because email offers no session to watch. A web page can be instrumented with scripts; a message in an inbox cannot. The only passive signal available is whether a resource hosted elsewhere was requested.

How an open is recorded

The mechanism is a single HTML image tag. The message body references an image on a remote server, conventionally a transparent one-pixel GIF. When the email client renders the message it requests that file, and the sending platform logs the request as an open.

What makes the request identifying is the URL. According to the CNIL recommendation adopted in March 2026, the embedded address carries parameters tied to the individual recipient: a pixel identifier, a campaign identifier and metadata. Fetching the image transmits the recipient's internet protocol address, a timestamp and a user agent string naming the client and operating system. That last field is how vendors reconstruct email client market share.

Two consequences follow. No fetch means no open, so a plain-text message or a client set to block remote content produces a delivered, read email logged as unopened. And the sender cannot tell a person from a machine, since both make the same request. Most platforms hedge with an implied open, crediting anyone who clicks a tracked link.

The count is also split two ways. Total opens records every fetch, including a subscriber returning three times; unique opens records one per recipient. Benchmark tables almost always use unique opens, while dashboards often show both without labeling which is which.

The arithmetic

The formula is stable across vendors. Germany's digital economy association BVDW, in a guide published on March 18, 2026, sets it out as unique opens divided by recipients minus bounces, multiplied by 100, and Mailchimp documents the same construction.

Removing bounces matters because an address that rejected the message never had a chance to open it. That ties the metric to list quality: the denominator is what remains after hard bounces are subtracted, and unprocessed bounces are how spam traps enter legitimate lists.

The subtraction is less precise than it looks. Delivered is inferred from the absence of a bounce, not confirmed by inbox placement, so a message filed as spam counts as delivered and drags the rate down without saying why. Open rate mixes three questions into one figure: did the message arrive, did it land somewhere visible, did anyone care.

Origin and evolution

The technique predates email marketing as an industry. Richard M. Smith published a Web Bug FAQ on November 11, 1999, later archived by the Electronic Frontier Foundation, describing a graphic "designed to monitor who is reading the Web page or Email message". It reproduced examples pulled from commercial mail, each a one-pixel image tag with the recipient's address in the query string. The architecture has not changed since.

The first structural disruption came from Google. In December 2013 Gmail began routing embedded images through its own proxy and serving cached copies, with automatic display rolled out in stages from December 12. The effects pulled in opposite directions. Opens rose, because recipients no longer had to approve images. Detail collapsed, because the request now came from Google: geolocation resolved to the proxy, browser detection failed, and the cache suppressed repeat opens.

The second disruption was larger. Apple previewed Mail Privacy Protection at its Worldwide Developers Conference on June 7, 2021 and shipped it with iOS 15 on September 20, 2021, followed by iPadOS 15, macOS Monterey and watchOS 8. According to Apple's legal documentation, Mail "downloads remote content in the background by default" whether or not the recipient engages, routing it through two relays run by separate entities so neither holds both address and content.

The design deliberately severs the link between a fetch and a human. Every message reaching a subscriber with the setting enabled registers as opened, at a time chosen by Apple and from an approximate region rather than a location.

What the numbers look like now

Scale determines the damage. Litmus, now part of Validity, measured email client market share across more than a billion opens in July 2026 and put Apple at 62.26 percent, Gmail at 27.03 percent, Outlook at 5.83 percent and Yahoo Mail at 2.59 percent. Opens from recipients using Mail Privacy Protection pass along no client version or platform detail and are not counted as reliable, and Litmus estimates the feature touches roughly 55 to 60 percent of all opens.

Published benchmarks carry the inflation. Mailchimp reports an average of 35.63 percent across all users, with non-profits at 40.04 percent and ecommerce at 29.81 percent. The data was last updated in December 2023, and the page warns that accuracy may be affected by Apple's privacy changes.

Figures conflict even within one vendor. The Mailchimp table gives 35.63 percent while the questions below it recommend aiming for 34.23 percent, and a separate Mailchimp reporting guide states the all-industry average is 21.33 percent, closer to the pre-2021 consensus. No reconciliation is published. Series spanning September 2021 are not continuous, and comparisons against tables of unknown vintage compare different measurements.

Why the metric survives its own unreliability

Diagnostic value outlived predictive value. A sudden collapse on a stable list is one of the few early warnings that mail is being filtered, because mailbox providers do not report inbox placement.

Deliverability rules reinforce the dependency. Google's requirements for bulk senders took effect on February 1, 2024 and require a spam rate below 0.3 percent in Postmaster Tools alongside SPF, DKIM and DMARC authentication. Complaints come from sending to people who do not want the mail, and the cheapest proxy for wanting it is engagement. Suppressing subscribers who have not opened in months is how senders hold complaints down, leaving an unreliable metric to govern list retention.

Money runs through it too. Newsletter advertising is priced against engagement figures that originate in open reporting, and email inventory carries measurement constraints that web inventory does not.

Limitations and disputes

The core objection is that the numerator counts events no human caused. Automated opens come from Apple's prefetching, from corporate security gateways that render images before delivery, and from filtering systems that open messages to classify them. None separates cleanly from a real read.

Vendors sell adjustments. Several platforms publish a machine-excluded open rate derived from timing patterns, user agent characteristics and known proxy address ranges. Those methods are proprietary, differ between suppliers, and carry no accreditation from the Media Rating Council. Two platforms measuring one campaign can report materially different corrected figures, and neither can be verified. Caching cuts the other way: Gmail and Apple both serve cached copies, so repeat engagement is undercounted.

A structural limit sits underneath, and vendors selling identity resolution press on it: open rate describes a known list, not a market, and email platforms cannot identify up to 95 percent of website traffic.

France moved the argument from methodology to law. The CNIL adopted its recommendation on tracking pixels in emails on March 12, 2026, published as deliberation number 2026-042 on April 14. It classifies the pixel as a tracker under Article 82 of the French Data Protection Act, the transposition of Article 5(3) of the ePrivacy Directive.

Analyzing open rates to measure and optimize campaign performance falls in the consent-required category, as does profiling recipients by expressed interest and detecting suspected fraud. A narrow exemption covers measurement used strictly to adjust send frequency or stop sending to inactive addresses, and even there retention is limited to the date, without the time, of the last known opening. Consent for the pixel is independent of consent for the message, so a transactional email that needs no consent to send may still need consent to track.

The text followed a public consultation in 2025 and arrived alongside parallel French action on session replay tools.

Disambiguation

Click rate divides unique clicks by delivered messages. It shares the denominator but requires a deliberate action, which is why it survived 2021 largely intact.

Click-to-open rate divides clicks by opens. Its denominator is the compromised number, so it inherits every distortion in open rate and amplifies it: machine opens push the denominator up and the ratio down.

Delivery rate and inbox placement rate are distinct. Delivery rate is messages accepted minus bounces. Inbox placement is the share reaching the inbox rather than the spam folder, and can only be estimated through seed lists or panels.

Open rate in direct mail describes physical envelopes opened by recipients, measured by survey. The phrase is identical; the methodology shares nothing with email.

Recent developments

The CNIL published a questions and answers document on July 22, 2026. Holders of addresses collected before April 14, 2026 had until July 14 to inform recipients and offer a means of objection. Senders who missed that notice must now obtain consent where it is required or stop using the pixels that need it, in business-to-business contexts as well as consumer ones.

Measurement is meanwhile being asked to describe a new kind of reader. The BVDW guide introduces an AI Agent Approval Rate, the share of messages an agent processes, analyzes or forwards, alongside a Meaningful Interaction Rate covering how often an agent acts in the sender's interest. Both accept that a growing share of message handling is machine handling. Open rate has been measuring exactly that since 2021 without admitting it.

Timeline

  • November 11, 1999: Richard M. Smith publishes the Web Bug FAQ, documenting one-pixel image tags used to track email reads
  • December 12, 2013: Gmail begins a staged rollout of automatic image display, serving embedded images through Google proxy servers and caching them
  • June 7, 2021: Apple previews Mail Privacy Protection at its Worldwide Developers Conference
  • September 20, 2021: iOS 15 ships with Mail Privacy Protection, followed by iPadOS 15, macOS Monterey and watchOS 8
  • December 2023: Mailchimp last updates the data behind its published industry open rate benchmarks
  • February 1, 2024: Google's bulk sender requirements take effect, including a spam rate ceiling of 0.3 percent in Postmaster Tools
  • October 7, 2024: The European Data Protection Board adopts Guidelines 2/2023, confirming that Article 5(3) of the ePrivacy Directive applies to tracking pixels in email
  • 2025: The CNIL runs a public consultation on a draft recommendation for email tracking pixels
  • March 12, 2026: The CNIL adopts deliberation number 2026-042 on tracking pixels in emails
  • March 18, 2026: BVDW publishes an email marketing guide setting out open rate formulas and new agent-oriented metrics
  • April 14, 2026: The recommendation is published, starting a three-month transition period
  • July 14, 2026: The transition period for previously collected addresses closes
  • July 22, 2026: The CNIL publishes a questions and answers document on the recommendation
  • July 2026: Litmus measures Apple at 62.26 percent of email opens across more than a billion tracked opens

Summary

Who. Senders and their email service providers calculate open rate; mailbox providers including Apple, Google, Microsoft and Yahoo determine how much of it is real; measurement firms such as Litmus and Validity report client-level distribution; and the CNIL now regulates the underlying tracking in France.

What. Open rate is unique opens divided by delivered messages, expressed as a percentage. The open event is a fetch of a remote one-pixel image whose URL identifies the recipient and the campaign, supplemented on most platforms by opens implied from clicks.

When. The tracking pixel was documented in email by November 1999. Gmail's image proxy changed what the fetch revealed in December 2013. Apple's Mail Privacy Protection broke the link between fetch and reader in September 2021. Consent obligations took hold in France through 2026.

Where. Measurement happens on the sending platform's servers, not in the inbox. Apple accounted for 62.26 percent of tracked opens in July 2026, which makes one vendor's default setting the dominant influence on a global metric.

Why. No better passive signal exists. Mailbox providers do not report inbox placement, complaint thresholds punish sending to disengaged addresses, and engagement is the cheapest available proxy for consent to keep sending. Open rate remains in use because it is the only number available, not because it is accurate.