Skip to content
All articles

Deliverability · 10 November 2026 · 7 min read

Apple Mail Privacy Protection and honest open rates

Rampsend team

An open isn’t a read

For years the open rate was the first number anyone looked at after a send. Then Apple changed how its Mail app loads images, and for a large share of most lists an "open" stopped meaning that a person had read anything. If your open rates jumped in late 2021 and never came back down, this is why. The fix isn't to throw opens away entirely; it's to know which opens are real and to lean on metrics that a proxy server can't fake.

How Mail Privacy Protection works

Apple announced Mail Privacy Protection (MPP) in June 2021 and shipped it with iOS 15 and iPadOS 15 in September 2021, followed by macOS Monterey. When someone opens the Mail app after updating, Apple asks whether they want to protect their Mail activity. Most people say yes.

With MPP switched on, two things change:

  1. Remote content is loaded privately in the background. Apple's servers download the images in a message, including any one-pixel tracking image, when the email arrives, not when the person reads it. Apple says this happens even if the message is never opened.
  2. The recipient's IP address is hidden. The image request comes from an Apple proxy, so the sender sees an approximate region at best, not the person's real location or network.

MPP applies to the Apple Mail app, not to the mailbox provider. Someone with a Gmail or Outlook address who reads it in Apple Mail on an iPhone is covered. Someone with an iCloud address who reads it in another app isn't, at least not by MPP.

Why open rates became inflated

Open tracking works by embedding a tiny image with a unique URL in the HTML. When the image is requested, the sending platform records an open. The method was always approximate, because it missed people who block images, but it failed in a predictable direction: it under-counted.

MPP flips that. Every message delivered to an MPP-enabled Apple Mail inbox can trigger a pixel fetch shortly after delivery, so it registers as opened. The result:

  • Open rates rise sharply on lists with many Apple Mail users, which describes most consumer lists.
  • Open times are wrong. The recorded time is when Apple fetched the image, often seconds after delivery, not when a human looked.
  • Location and device data degrade. The IP belongs to a proxy, so open-based geography is unreliable for those recipients.
  • Anything built on opens breaks quietly. Subject line tests decided by open rate, "send time optimisation" based on open times, re-engagement segments of "people who haven't opened in 90 days" and automations that wait for an open all start acting on machine behaviour.

That last group is the costly one. A sunset policy that removes non-openers will never remove an inactive Apple Mail user, because their client "opens" everything. Meanwhile a subject line test may crown whichever variant happened to land in more Apple inboxes.

How machine opens are detected

You can't stop the prefetch, but you can usually recognise it. Detection relies on the fact that a proxy behaves differently from a person:

  • Where the request comes from. Prefetch requests arrive from Apple's proxy infrastructure, and the request headers look consistent across recipients in a way that real device requests don't.
  • When it happens. A fetch that lands within moments of delivery, for every message to that recipient, regardless of time of day, looks automated.
  • What happens next. A machine open is never followed by a click from that open. A real reader sometimes clicks, replies or comes back later.

No method is perfect, and a machine open can hide a real one: if Apple prefetched the image, a later human read may not produce a second request you can see. So treat the split as two honest numbers, not one precise one. "Opens from people, as far as we can tell" and "opens from machines" are both useful; mixing them is not.

Rampsend detects machine opens and reports them apart from real opens, in the dashboard and in the get_stats tool of the MCP server, so your headline open rate isn't padded by Apple's proxies.

Metrics to trust instead

The better metrics share one property: they require a person to do something.

MetricWhat it tells youCaveat
ClicksSomeone found the content worth acting onSome corporate security tools follow links to scan them, so a burst of instant clicks can be automated too
RepliesStrong engagement, and a positive signal to mailbox providersOnly meaningful if your From address accepts replies and someone reads them
ConversionsThe outcome you actually wanted: a purchase, a sign-up, a feature usedNeeds attribution, such as tagged links or events tracked in your product
Unsubscribes and complaintsWhether the email was unwelcomeLagging and small in absolute numbers, so watch the trend
Human opensA rough, directional sense of attentionUnder-counts Apple Mail readers; compare it only with itself over time

Clicks and click-to-open

Click rate (clicks divided by delivered) is now the main engagement number for most senders. Click-to-open rate is only meaningful if the "open" part excludes machine opens; otherwise the denominator is inflated and the ratio falls for no real reason.

Conversions through events

The most reliable measure is what people do after the email. If you track product events, such as a purchase or a trial start, you can compare the people who received a campaign with what they did next. Rampsend's events API records actions like Placed Order against a profile, so you can build segments from behaviour rather than from opens.

Engagement for list hygiene

For sunset and re-engagement rules, base "engaged" on clicks, replies, purchases or sign-ins over a window, not on opens. You'll remove more genuinely inactive contacts, which lowers bounce and complaint risk; see bounce and complaint rates for why that matters.

Fixing tests and automations that rely on opens

A short audit:

  1. Subject line tests. Decide winners on clicks or conversions. If you must use opens, use human opens only and accept a smaller signal.
  2. Send-time optimisation. Don't learn send times from Apple-proxied opens; their timestamps reflect delivery, not reading.
  3. Automations that wait for an open. Change "if opened, then…" branches to "if clicked" or to a product event.
  4. Reports to stakeholders. Show real opens and machine opens separately, and explain the change once, so nobody celebrates a jump that came from a privacy feature.

Turning tracking off for sensitive mail

Open and click tracking aren't free. A tracking pixel is a remote request, and click tracking rewrites every link through a redirect. For some messages that's the wrong trade:

  • Password resets and sign-in links. A rewritten link adds a hop, can look suspicious and gains you nothing.
  • Security alerts, legal or account notices. You care that they were delivered, not whether they were opened.
  • Messages to audiences who expect privacy, such as health, finance or support conversations.

On Rampsend, tracking is set per message. Pass track_opens: false to skip the pixel and track_clicks: false to leave links untouched:

BASH
curl https://app.rampsend.com/api/v1/emails \  -H "Authorization: Bearer rs_live_…" \  -H "Content-Type: application/json" \  -d '{    "from": "Acme <security@mail.acme.com>",    "to": "ada@example.com",    "subject": "Your sign-in link",    "html": "<p><a href=\"https://acme.com/login/…\">Sign in to Acme</a></p>",    "track_opens": false,    "track_clicks": false  }'

Delivery status still works without tracking. GET /api/v1/emails/:id reports whether the message was delivered, bounced or drew a complaint, which is what matters for transactional mail. Transactional vs marketing email covers the other ways those two streams should differ.

What to report from now on

Keep opens, but demote them. A sensible dashboard for each send shows delivered, bounces, complaints, unsubscribes, human opens, machine opens and clicks, with conversions alongside when you track them. Judge a campaign by clicks and outcomes, use human opens as a directional check, and treat machine opens as a reminder of how many of your readers use Apple Mail, not as attention.

If you want to see the split on your own list, Rampsend's Free plan covers 1,000 contacts and 6,000 emails a month; the quickstart gets you sending in a few minutes.

Frequently asked questions

Frequently asked questions

When did Apple Mail Privacy Protection start?

Apple announced it in June 2021 and shipped it with iOS 15 and iPadOS 15 in September 2021, followed by macOS Monterey. Users are asked whether to turn it on when they first open Mail after updating.

Does MPP affect Gmail addresses?

Yes, if the person reads their Gmail account in the Apple Mail app with MPP turned on. MPP depends on the email app, not the mailbox provider.

Are open rates useless now?

Not entirely. Human opens, with machine opens removed, are still a useful directional signal when compared with themselves over time. They just shouldn't decide tests, automations or list clean-up on their own.

What is a machine open?

An open recorded because software, such as Apple's privacy proxy, fetched the tracking image rather than a person viewing the email. Rampsend reports machine opens apart from real opens.

How do I turn off tracking for one email?

On Rampsend, set track_opens: false and track_clicks: false on the send request. Delivery status is still reported. See the API reference.