Your tracker says the email was opened. Should you believe it?
The honest answer: it depends on the open. Some opens are a prospect reading your proposal at their desk. Others are Apple's servers pre-loading images, or a security scanner checking your links for malware. Treating all of them as equally true is how people end up chasing ghosts. This guide is about the right amount of trust, at each level.
What does confidence mean for an email open?
It means treating an open as a probability instead of a fact. The useful question isn't "did they open it," it's "how likely is it that a real person opened it."
That's a shift in thinking, but it matches how the signal actually works. An open is recorded when a tracking image loads, and lots of things load images: people, privacy features, scanners, proxies. So every open arrives with some amount of doubt attached. Good tracking measures that doubt and shows it to you. Bad tracking hides it behind a confident-looking number. There's more on what the base metric can and can't tell you in our glossary entry on [email open rate](/glossary/email-open-rate).
Why can't trackers just count opens?
Because most of what they'd count isn't people. Apple Mail Privacy Protection pre-fetches tracking pixels on Apple's servers for every recipient, read or not, and Apple Mail holds roughly 58 percent of email client share per Litmus.
Then come the scanners. 15 to 25 percent of B2B recipients sit behind corporate security tools like Mimecast, Proofpoint, or Microsoft Defender, which open every link and image at delivery to check for malware. One scanner can fire many opens within seconds. Gmail adds a third wrinkle by serving images through its own proxy, which makes those opens harder to judge on sight. Add it up and raw open rates on B2B lists run 2 to 3 times inflated. A reported 70 percent open rate is often 25 to 35 percent real reads.
How does tier grading work?
Every open gets a grade from Tier 1, high-confidence human, down to Tier 5, bot or scanner. The grade comes from request timing, network origin, and client signature.
Timing asks: did this open land at a human moment, or two seconds after send? Origin asks: does this request come from where a reader would be? Signature asks: does the client look like a mail app a person uses, or like a known scanner? No single signal decides alone. It's the combination that separates a Tuesday-morning read from a delivery-time malware check. Opens that score below a 25 percent confidence floor are excluded from your stats entirely, because a wrong score is worse than no score. The full logic is public on our [methodology page](/methodology), with the tier details on the [confidence scoring](/features/confidence-scoring) page.
How accurate is the grading?
The model agrees with human-rated classifications 95 to 98 percent of the time. Tier 1 false positives, machine opens wrongly graded as high-confidence human, run under 2 percent.
And you can audit it yourself. The [Diagnose view](/features/diagnose-open-evidence) lets you click any open and see why it counted or got filtered, in plain words. If a grade ever looks wrong to you, the evidence is right there to check. That's the standard worth holding any tracker to: not "trust us," but "here's why."
What should you do at each confidence level?
Act on the top, wait on the middle, ignore the bottom. Tier 1 and 2 opens are follow-up material. Tier 3 means hold until a second signal shows up, like a repeat open or a click. Tier 4 and 5 are machine noise.
This maps cleanly onto real decisions. High confidence at a human hour: send the follow-up while you're on their mind. Ambiguous mid-tier open: do nothing yet, one more signal usually arrives within a day if the interest is real. Scanner tier: pretend it never happened, because for your purposes it didn't.
Does this change how you follow up?
Yes, in one big way: you stop spending energy on ghosts. Every hour spent chasing scanner opens is an hour not spent on the prospect who read your email three times this week.
If your current tool hands you one undifferentiated count, our [comparison hub](/compare) shows how the major trackers handle machine noise. Or test it directly: Outsolvi is $7 a month billed yearly, $12 monthly, with a 14-day free trial, no credit card, and a 7-day money-back guarantee. Two weeks of graded opens next to your old raw counts will tell you exactly how much of your open data was ever real.