MCP

Tracking caveats

Last verified: 2026-08-05

TL;DR

  • Open tracking has been structurally broken since 2021. Apple's Mail Privacy Protection pre-loads images, and Apple Mail holds roughly half the global client market — so a large share of "opens" are machines (2026) [1].
  • Link tracking has a milder version of the same problem: security scanners click links before a human sees them.
  • The pixel is not free. It is a signal filters read, and there is a credible school that says remove it entirely.

How the pixel broke

Open tracking works by embedding an invisible image; when a mail client requests it, the send is recorded as an open. Apple's Mail Privacy Protection, introduced with iOS 15 in September 2021, pre-loads that image regardless of whether anyone reads the message [1]. Corporate security gateways do the same when scanning attachments and images, and so do preview panes.

The result is not a small bias but a large and uneven one — it varies by how much of your list uses Apple Mail and how aggressive their employers' scanners are, which means it differs between segments and therefore corrupts comparisons between them. Published open rates now span 27.7-44% depending on whose corpus you read and how much inflation is baked in [2].

Note what this does to the classic diagnosis. "High opens, low replies means the copy is the problem" was reasonable when opens meant reading. Now an inflated open rate can coexist with mail that no human saw, so opens can no longer clear delivery from suspicion — Diagnosing underperformance.

Automated scanners follow links to check them, producing clicks with no human behind them. Click-through data is more reliable than opens because scanning is less universal than image pre-loading, but a click still is not a person until something else confirms it.

Tracked links carry a second cost: rewriting a URL to route through a tracking domain is one of the patterns filters associate with bulk mail, particularly when the tracking domain is shared with other senders.

The pixel's cost to deliverability

Beyond measurement error, there is a live argument that open tracking hurts placement — a shared tracking domain and an invisible image are both bulk-mail signals, and practitioners report placement improving when tracking is switched off. The opposing school treats tracking as a workflow signal worth the cost: knowing who engaged drives prioritization, even if the number is noisy.

Fits when (tracking on): small lists where a human reviews each prospect and the ranking is genuinely used. · Avoid when: high volume, cold lists, or any campaign where deliverability is already marginal.

Fits when (tracking off): most cold outreach at volume — replies are the metric that matters, and one bulk signal is removed. · Avoid when: you have no other way to prioritize a large warm list.

Default if the user has no preference: tracking off for cold campaigns. The measurement it buys is corrupted anyway, and the cost is real (Deliverability in a nutshell).

What to trust instead

Replies. A reply is a human act with intent behind it and no automated analogue. The recommended substitution is direct: reply rate is the only signal that confirms a real person read the message and cared enough to respond [1].

Bounces, which are protocol-level facts rather than inferences.

Meetings and revenue — sparse, delayed, and the only measures of the thing you wanted.

If you keep open tracking, treat it as a coarse relative signal within a single campaign and never as an absolute number, never across segments, and never in a report to anyone who will act on it.

References

  1. Smartlead — Cold email open rate accuracy (vendor blog, Jun 2026)
  2. Woodpecker — Cold email statistics (20M+ email corpus, updated Jun 2026)