For roughly the first decade of B2B email tracking, call it 2010 to 2020. the math was simple. A tracking pixel loaded when the recipient opened the email. The dashboard counted that as an open. The number was directionally correct, the rep acted on it, and the system worked.
That system is now broken. By a lot. This piece is the documented version of what changed, what it costs sales teams, and what to measure instead.
The four forces that broke open tracking
Apple Mail Privacy Protection (since 2021)
Apple shipped Mail Privacy Protection in iOS 15 in September 2021[1]. The behavior: any image embedded in an email, including the tracking pixel, gets pre-fetched by Apple's servers when the email is delivered, regardless of whether the recipient ever opens it.
The mechanism is straightforward. Apple downloads the email content to a privacy-preserving proxy, fetches all remote images server-side, caches them, and serves the cached version when the user actually opens the message. From the sender's tracker's perspective, every email to an Apple user looks "opened" within minutes of send, from an Apple IP, with no way to distinguish the proxy fetch from a real human read.
The reach: Apple Mail accounted for 62% of the opens Litmus measured in July 2026, across all the senders it tracks[2], and Litmus says studies now put MPP at roughly 55 to 60 percent of all email opens. On a list that reads mostly on iPhones and Macs, this one source can account for most of your reported opens.
Gmail's image proxy (since 2013)
Google routes every image in Gmail through its own image proxy, a change it announced in December 2013[3]. When a recipient opens your email in Gmail, your server doesn't see their IP address or device. It sees Google's proxy.
Unlike Apple's, this fetch happens when the email is viewed, so a Gmail proxy open is usually a real open. The distortion is in the counting. One view can reach your tracker as several fetches in quick succession, and a later view may send none if Gmail shows a cached copy. A tracker that counts every fetch overstates Gmail opens, and one that throws proxy fetches away misses real readers. The Gmail image proxy write-up has the evidence.
Corporate email scanners (always there, now mandatory in many environments)
Microsoft Defender for Office 365, Proofpoint, Mimecast and others check the links and images in incoming email[4]. The purpose is security: malware hidden in images, phishing redirects, known-bad domains. The side effect on tracking is dramatic.
An enterprise email-security stack can open your tracking pixel within seconds of delivery, from the company's own network, and click every link while it's there. The buyer might open the email an hour later, by which time the scanner has already fired. A dashboard that counts both reports mostly the scanner.
Omnivery found that at large corporate domains, 80 to 95 percent of clicks came from these scanners rather than people[5]. For lists heavy in financial services, healthcare and large enterprises, that makes scanners the biggest source of noise in your click data and a large one in your opens.
"Self-opens" (the sender re-reading their own sent items)
This one is the smallest of the four but the most consistently underappreciated. Every time you open your own sent email, to remind yourself of the wording before a call, to forward it to a colleague, to copy-paste a line into a follow-up, the tracking pixel loads. Most trackers count that as an open.
For an AE running a long deal cycle, self-opens pile up on exactly the deals that matter most. The proposal you keep re-reading. The follow-up you're drafting. Each load is a phantom "open" that doesn't belong in your stats.
What this costs sales teams
The cost of the noise is not the noisy number itself. Teams report 70% open rates and 2% reply rates and intuitively understand the two don't reconcile. The real cost is the downstream decisions made on the noise.
Reps chase dead threads
The most common pattern: a tracker fires "5 opens on the proposal" alert. The rep sees five opens in two days and assumes high interest. They call. The buyer doesn't pick up. They email a "just checking in" follow-up. Still nothing. They call again, follow up again, and burn three hours of attention on a thread where the "five opens" were the buyer's iPhone pre-fetching the email five times because of MPP.
Multiply that pattern across every active thread on a rep's list and the wasted hours add up fast, all of them spent on engagement that was never real.
Reps miss the actually-warm threads
The mirror problem: the rep stops trusting opens entirely. When a thread genuinely warms up, a buyer reading the proposal three times in 48 hours, distinct IPs from the buyer's office and their home, it looks the same in the dashboard as the noise. The rep tunes out. The warm thread goes cold because the rep didn't act on the signal that was actually real.
This is the harder cost to attribute, because you only notice it on the deals that closed elsewhere or stalled out. It is just as real.
Cadence and routing logic break
Most modern sales-engagement platforms run rule-based routing: "if a prospect opens 3+ times in 48 hours, route to the rep's hot-lead queue." Those rules were designed when opens were directionally accurate. Fed raw opens in 2026, those same rules misroute a lot of the queue.
The fix is either to filter the opens before they hit the rule, or to rewrite the rules around different inputs (replies, clicks, multi-channel engagement). Most teams haven't done either.
What to measure instead
The honest answer for B2B teams in 2026:
Reply rate is the cleanest single signal. No proxy generates a reply. No scanner sends a polite "thanks, not interested." Reply rate survives the entire noise stack intact. A B2B team that anchors on reply rate, weighted by sentiment, will make better decisions than one anchored on opens, even if the open data improves.
Clicks on high-intent links come second, with a caveat. Security tools click links too: at large corporate domains most clicks are scanners, and Omnivery saw the bot share of Gmail clicks jump to about 22% in early 2026[5]. Apple's MPP loads images, not links. A click from a person on your pricing or comparison page is still a strong intent signal; the job is telling it apart from a scanner's.
Confidence-scored opens are the third piece. This is where Outsolvi sits. Every open is graded as it arrives, from its user agent, network, timing and the pattern of requests around it: verified human, most likely human, possibly human, or a machine such as Apple's preview or a security scanner. Only the first two count. The dashboard reports that number, not the raw count, and shows the reason next to each open.
We expect grading like this to become normal across trackers, because the noise problem is now well understood.
What to do about it this quarter
Three things in priority order, if you're running B2B outbound today:
1. Stop reporting raw open rate as a single number to leadership. On many lists it's mostly machines. Replace it with reply rate plus a "high-confidence opens" sub-metric if your tracker exposes one. Leadership will adjust. The expectations math gets easier, not harder, because the new metrics are defensible.
2. Rewrite hot-lead routing rules to weight replies and high-intent clicks over open counts. A prospect who replies to a touch within 4 hours is qualitatively different from one who "opened" five times. Route on the former; deprioritize the latter.
3. Either switch to a tracker that confidence-tiers opens, or build the filter yourself on top of whatever you have today. This is the structural fix. Vendors that don't tier are reporting numbers that no rep should be acting on. Vendors that do tier are surfacing the cleaner signal. The third option, staying with a non-tiering tracker and ignoring the open dashboard, also works but wastes whatever you're paying for the tracker.
Where the numbers in this piece come from
Every figure here comes from a published source, linked in the notes: Litmus for Apple Mail's share of opens, Apple for when Mail Privacy Protection shipped, Google for the Gmail image proxy, Microsoft for Defender's link scanning, and Omnivery's 2026 bot report for measured machine opens and clicks.
An earlier version of this report quoted Outsolvi's own figures: a 30 to 50 point inflation range, per-industry breakdowns and a follow-up conversion lift. We don't have data behind those, so they're gone. Outsolvi's methodology page explains how opens are graded. Every rule is checked weekly against opens that a reply later confirmed or a user corrected.
What's not in this report
We deliberately don't make claims about specific competitor accuracy. That gets adversarial fast and most of the data we'd cite isn't ours to share. Our comparison pages handle the head-to-head competitor analysis with direct documented references. This piece is about the industry-wide pattern, not which vendor is best.
We also don't claim Outsolvi solves the noise problem perfectly. The noise keeps changing (Apple ships new MPP behavior, Google adjusts its proxy, Microsoft updates Defender), so every grading rule is checked weekly against opens that a reply later confirmed or a user corrected, and our team adjusts the rules that slip. This piece was last reviewed in September 2026.
If you want to estimate the noise on your own send patterns, we built a free calculator that takes your list size and audience composition and outputs an inflation range. No signup required.
The takeaway
The 2010-2020 generation of email tracking measured what mailbox-provider behavior of that era let it measure. The 2026 generation has to measure around mailbox-provider behavior that was specifically designed to break the old pattern.
Teams that recognize the shift, change what they measure, and pick tools that survive the new measurement environment will run cleaner pipeline. Teams that keep optimizing on raw open count will spend the next 24 months wondering why the funnel doesn't respond the way it used to.
If you want our take on the cleanest way through, try Outsolvi free for 14 days. The default dashboard filters proxy noise and surfaces the actual human reads. The machine opens stay visible on each email with the reason they were set aside, if you ever want the raw picture for diagnostic purposes.