Predict your cold email open rate.

Apple's own documentation for Mail Privacy Protection describes the mechanism plainly. When a message reaches Mail on an Apple device or iCloud.com, the client downloads all remote content, including the tracking pixel that reports an open, in the background and by default, whether or not anyone actually reads the email (source: Apple, Mail Privacy Protection and Privacy). The download is routed through two separate relays so no single party can tie the fetch to both a real IP address and the message content at once. The privacy design is good for the recipient. For a sender reading an open-rate report, it means a meaningful share of every open on file is really the client prefetching a pixel, not a person reading the subject line and deciding to open the message.

The scale of that effect is not small. Litmus's own market-share report, built from more than a billion tracked opens, lists Apple (Mail, iPhone, iPad, and Mail Privacy Protection combined) as the single largest email-client category tracked, and states that Apple's MPP now affects roughly 55 to 60 percent of all email opens it measures. Apple and Gmail together account for close to 90 percent of total market share, so almost every cold-email list built from public B2B contact data runs through one of the two platforms carrying the least reliable open signal in the inbox.

None of this makes the predictor above wrong. It means the number it returns describes what the FORKOFF dataset recorded as opens under the same tracking mechanics every open-rate benchmark on the internet uses, a fired tracking pixel, not a verified human glance. The predicted band already reflects that uncertainty: it stays wide, plus or minus 6 percentage points, because list composition alone cannot resolve how many of those opens were prefetches rather than reads, and the underlying dataset carries the same Apple-and-Gmail skew as every other benchmark cited on this site.

Two adjustments make the output more useful in practice. First, treat the top of the predicted band as a ceiling, not a target: a Web3 founder segment predicted at 60 percent almost certainly includes a real prefetch share once Apple Mail recipients enter the mix, since founders read mail on iPhones as often as anyone. Second, once a campaign has actually sent, read it against reply rate rather than open rate alone, because a reply requires a human to act, and Apple's background prefetch cannot manufacture one. That is the metric the next section runs the numbers on.

There is a simple tell for a prefetch-inflated open versus a genuine one, and it is worth watching on any list heavy with Apple Mail recipients. A real open clusters around when the recipient actually checks their inbox, often hours after delivery and skewed toward business hours. A prefetch fires within seconds of delivery, every time, regardless of time of day, because the client is not waiting for a human, it is loading remote content the moment the message lands. A campaign where the open curve spikes immediately at send time and then goes flat is not evidence of a great subject line. It is evidence of a list running mostly through Apple's relay, and it is a reason to weight that campaign's reply rate more heavily than its open rate when deciding whether the copy or the targeting needs work.

The prefetch effect is not evenly distributed across the industries this predictor already segments by, which matters when reading the industry delta table above. B2B SaaS buyers skew corporate, and a large share of corporate mail still runs through Outlook or Google Workspace rather than personal Apple Mail, so a SaaS list's reported open rate is closer to a real read-rate than the cross-industry average suggests, one reason the -8pp SaaS delta in the methodology table above is probably closer to the true gap than it looks. Web3 and DeFi recipients skew the opposite way: founder-heavy, mobile-first, and disproportionately on Apple hardware, so the +8pp Web3 delta is the segment most likely to be partly a prefetch effect rather than a pure engagement signal. Reading the predictor's industry deltas next to this split is more useful than reading either number alone.

Company size tells a similar story from a different angle. The predictor's steepest delta, enterprise recipients at -10 percentage points against 1-to-10-employee companies at +7, is not purely a difference in how interested the two audiences are. Enterprise inboxes are more likely to sit behind Microsoft 365 or Google Workspace with corporate mail-security scanning that pre-fetches links and images for threat detection, independent of Apple's own privacy layer, which means enterprise open data carries its own separate inflation source on top of everything above. Small, founder-run companies are more likely to read mail on a personal device with no corporate security layer in front of it, so their open numbers are, on balance, closer to a genuine read than the enterprise segment's are. None of this changes the predicted band. It changes how much weight to put on the top versus the bottom of it once the campaign is actually running.

The FORKOFF dataset recorded 17% of its replies as genuinely positive, the rest routine acknowledgments and declines. Cleverly's read of more than 100 million emails, Instantly's benchmark across billions of tracked interactions, and Mailshake's own dataset all put the honest cold-email reply-rate baseline at roughly 3.43%. One major sending platform states the gap even more directly from its own comparison tool: across 37,000 tracked campaigns the median open rate is 26% and the median reply rate is 1.5%, and that platform's own conclusion is that open rate is the noisier of the two metrics, so reply rate is the number worth treating as ground truth.

The reason the gap holds up is structural, not a quirk of any one dataset. An open can be recorded by a background process before a human ever sees the subject line. A reply cannot: it requires someone to read the message, form an opinion, and take an action that costs them a few seconds of typing. That is a much higher bar, which is exactly why the number sits so much lower, and exactly why it is a cleaner signal of whether the campaign is actually working. Lemlist, which refuses to publish an open-rate benchmark at all, uses reply rate as its whole rubric: above 5% is good, above 8% is excellent, and below 3% means something upstream is broken, whether that is targeting, subject-line specificity, or list quality. Instantly's data puts top-quartile campaigns at 5.5% reply and elite campaigns above 10%.

None of this replaces the predictor above. Open rate is still the earliest signal available, often visible within hours of a send, while reply rate needs the full follow-up window to read cleanly. The right sequence is to use the predicted open-rate band as a same-day sanity check on whether the send even reached inboxes and got looked at, then judge the campaign itself on reply rate once the sequence has run its course. A campaign that hits its predicted open band but posts a reply rate under 1% is not a subject-line problem: it is a targeting or offer problem the predictor was never built to catch, and it is worth reading the founder-funnel strategy for what actually closes that gap.

Cleverly's read of more than 100 million emails puts the average at 27.7%. Instantly's benchmark across billions of tracked interactions reports 44%. HubSpot's cross-industry figure sits at 42.35%. All three are measuring the same underlying activity, cold B2B outreach, and all three are correct for the population each one actually tracked. A benchmark built from senders who warm domains and send from named identities will always read higher than one built from a mixed pool that includes role addresses and cold domains, independent of how good anyone's subject lines are.

Geography moves the number by a similar margin, and for the same reason. One major sending platform's own data across 37,000 campaigns puts the median open rate at 38% in France, 36% in Germany, 32% in Poland, and 23 to 26% in the United States, depending on whether very high-volume senders are excluded from the US cohort. That spread tracks inbox-provider mix and regional sending infrastructure maturity far more than it tracks the quality of anyone's copy. A number below 21% is usually a deliverability problem worth fixing before touching a single word of the email; a number inside the 27% to 44% range is normal, and the levers below move it from the low end to the high end of that band.

Before Mail Privacy Protection shipped with iOS 15 in September 2021, a tracked open was a reasonable proxy for a human glance, and open rate alone was a defensible headline metric. Validity's own deliverability research found reported opens now running 18 to 32 percentage points above verified engagement, which is the gap MPP alone accounts for. Every benchmark published after 2021 is measuring a mix of real opens and background prefetches in proportions that shift as Apple's device share shifts, which is why a single-year benchmark table is already a moving target and why this predictor pairs its output with the reply-rate section above rather than reporting open rate alone.

The second shift is on the sending side, not the reading side. Starting February 1, 2024, Google requires every sender to Gmail personal accounts to authenticate with SPF or DKIM, keep valid forward and reverse DNS records, send over TLS, and hold the spam rate reported in Google Postmaster Tools under 0.3%, with Yahoo running an equivalent set of requirements the same year. Domains that do not meet the bar get rate-limited, blocked, or routed to spam before an open rate ever gets measured, which is exactly why domain warmup and sender identity sit at the top of the levers below: they are no longer best practice, they are the entry requirement Google and Yahoo now enforce directly.

None of the five levers above are exotic, and none of them require a bigger list or a cleverer subject line before the fundamentals are fixed. Warm the domain, send from a named person, clean the list, keep the subject short, and time the send and follow-up cadence deliberately, in that order. Do those five things and the predictor above stops being a ceiling estimate and starts describing something close to what the campaign will actually return. Skip them and no amount of subject-line testing buys back what an unwarmed, role-addressed, dirty list already gave away.