This is how Outsolvi decides whether an open was a person. It's written from the rules the product actually runs, in plain words, so you can judge the grading for yourself rather than take a score on trust. It's the technical companion to The State of B2B Email Tracking in 2026.
Why a raw open count stopped meaning much
A tracking pixel is a tiny image. When something loads it, the tracker records an open. For years that something was almost always a person's mail app. Three changes broke that.
Apple Mail Privacy Protection. With Protect Mail Activity on, Apple's Mail app downloads remote content in the background when a message arrives, whether or not the reader engages with it, and it does so through relays that hide the reader's address.[1] Apple Mail accounted for 62 percent of the opens Litmus measured in July 2026.[2]
Image proxies. Since 2013, Gmail has fetched images through Google's own servers rather than from the reader's device.[3] That hides the reader's location, but the fetch still happens when someone opens the message, which matters below.
Security scanners. Corporate email security checks links and images before people touch them. With Safe Links turned on, Microsoft Defender for Office 365 scans a message's links before delivering it,[4] and Proofpoint and Mimecast rewrite links so they can check each destination.[5][6] At large corporate domains, Omnivery found 80 to 95 percent of clicks came from bots.[8]
Put those together and a raw open count mixes real reads with machine traffic in proportions you can't see. The fix isn't a smaller number. It's grading every open and showing why.
The five grades
| Grade | What it means | Counted? | Example reason you'll see |
| Verified | A reply came back on the thread | Yes | reply received |
| Most-likely human | It looks like a person opening the email | Yes | opened in Gmail, opened in Outlook desktop |
| Possibly human | Too little to go on, or it could be you | No, shown | could be your own Sent folder, datacenter (could be corp egress) |
| Machine preview | Apple fetched the image on delivery | No, unless a click or reply follows | reached an Apple Mail inbox |
| Scanner | A security system or a script | No | Microsoft Defender security scan, faster than any human could open |
Behind the grades are four scores: verified 100, most-likely human 70, possibly human 20, and 0 for both machine grades. Anything at or above 25 counts. Your own opens are set aside with their own reasons, such as "your device, at send time".
Rule one: who fetched the image
The first question is whose computer asked for the pixel.
- Apple's relays mean Mail Privacy Protection. Apple publishes the address ranges they use, and Outsolvi refreshes that list daily and checks every open against it. The open is graded machine preview.
- Google's and Yahoo's image proxies mean the reader opened the message in Gmail or Yahoo Mail. The open counts. Google's proxy often fetches the same image several times for one view, from different addresses, so every fetch within ten minutes of a counted proxy open is grouped under it.
- Microsoft's cloud, minutes after delivery, is Microsoft Defender checking the message. Those hits are graded scanner.
- A datacenter is where scanners live, but it's also where many companies route their staff's traffic. A datacenter request with a stripped-down browser signature soon after delivery is graded scanner. One with a full browser signature could be a person behind a company gateway, so it's graded possibly human.
- An ordinary network with a full browser signature, more than a second after delivery, is graded most-likely human. Repeat views from the same device are grouped into one open.
Rule two: what identified itself
Every request carries a User-Agent, a short string naming the software that made it.[7] Some security scanners name themselves there, and those are graded scanner. Scripts and HTTP libraries don't read email, so a request from one is set aside as an automated client. A request that looks like Outlook for Windows fetching the image is labeled "opened in Outlook desktop".
Rule three: timing
Some timing is physically impossible for a person. An open that lands faster than anyone could open and render an email is graded scanner, as are bursts of opens arriving together, which is how scanners that sweep a whole mailbox behave. A scanner that opens the email and clicks its links in one pass shows up as an open-and-click combo and is set aside the same way.
Timing also catches you. An open from the IP address you sent from, in the first minute after sending, is your own device, often an add-in or a browser tab loading the image. And if you send from a gmail.com address, a Google proxy fetch in the first five minutes could be you checking your Sent folder, because that view goes through the same proxy. It's graded possibly human rather than counted.
Rule four: what happened next
The strongest evidence arrives later. A reply on the thread upgrades the open to verified. A click on one of your links after an open supports it. An Apple Mail fetch isn't counted as an open; if that reader clicks or replies, the click or reply is recorded on its own.
Clicks get the same treatment as opens. When a link is pasted into Slack, Discord, LinkedIn, WhatsApp, Telegram, Facebook or Twitter, the app fetches it to build a preview, and that's labeled as a link preview in that app, not a click by your prospect.
Rule five: was it you?
A tracker that counts your own opens flatters you. Outsolvi sets aside an open when:
- your Outlook add-in or Gmail extension told Outsolvi you were the one opening it
- it came from an IP address you've told it to ignore, or a device you marked with Not me
- you sent the email to yourself
- it came from the exact IPv6 address you sent from, which in practice only your own device has
- it came from a device you've signed in to Outsolvi from
It deliberately doesn't treat a shared office IPv4 address as you. A whole office can share one address, and a colleague of your prospect behind the same kind of network is a real reader.
Where the rules can still be wrong
- A reader in Apple Mail who never clicks or replies is never counted, even if they read every word. The rules would rather miss a read than count a machine.
- A person behind a corporate gateway can look like a datacenter and land in possibly human.
- Someone on your own office network opening the email in Outlook can look like you re-reading it. That's graded possibly human with the reason "Outlook on your network, could be you re-reading".
These cases stay visible on the timeline with their reasons, so you can make the call yourself.
How we check the rules
There's no single accuracy percentage, and we don't quote one. Every week, for each rule that fired at least five times in the last 30 days, Outsolvi compares how often a reply or a user confirmed it with how often a user overrode it. A rule that starts misfiring shows up as its own number falling, and our team changes it. The full method is on the methodology page.
What to do with the grades
Act on opens that count, clicks that follow them and replies. In Outsolvi, an email becomes a hot lead once it has five or more real opens and two or more link clicks, and you see a flame on it. Don't mention in your follow-up that you saw someone open the email. Use the signal to decide when to write, not what to write.
If you want to see how your own opens break down, the 14-day trial needs no card, and every open on every tracked email shows its grade and the reason from the first send.