What HumanOpens filters

Every click and open SendGrid records gets checked against the systems that fake engagement: security gateways, privacy proxies, and inbox image fetchers. No single clue decides. HumanOpens weighs timing, browser fingerprints, networks, and behavior across your whole send, then labels each event a person, a machine, or too close to call.

  • 8 kinds of machine traffic
  • 17 checks on every event
  • 5 guardrails that protect real readers

Who's clicking and opening besides people

These run at your recipients' companies and inbox providers. They behave differently, so each needs its own fingerprint.

SystemWhat it doesHow it shows up in SendGrid
Microsoft Defender for Office 365Opens mail and follows links in a sandbox before delivery, then again when Safe Links checks a click.Opens and clicks within seconds of delivery from Microsoft cloud networks, using one fixed, out-of-date browser signature at hundreds of unrelated companies.
MimecastRewrites and pre-scans links as mail arrives.Clicks on many links in one email, close together, from gateway networks shared by many tenants.
ProofpointURL Defense rewrites links and sandboxes the destinations.Clicks on every link, unsubscribe included, seconds after delivery, often with legacy browser signatures.
Barracuda, Cisco Secure Email, Trend Micro, Sophos, and other gatewaysScan attachments and links on arrival, sometimes in several passes.Repeated bursts of clicks on one email, each pass with a different device and browser.
Apple Mail Privacy ProtectionDownloads every email's images when it arrives, whether or not anyone reads it.An open for nearly every Apple Mail recipient, from Apple's proxy networks, with almost no browser detail.
Gmail image proxyFetches images through Google's servers. Usually when someone views the email, sometimes on delivery.Opens from Google's proxy with the same signature for every Gmail reader. Some are people, some aren't, which is why timing matters.
Yahoo and AOL image proxiesRoute images through their own servers.Opens that hide the reader's device and location behind the provider's network.
Cloud and data center trafficLink checkers, crawlers, and monitoring tools running on hosting providers.Clicks from server networks where no person reads email.

The checks

Grouped by the kind of evidence they use. Each check adds or removes weight. An event is called a machine only when the evidence adds up.

Timing

  • Seconds after delivery. Opens and clicks that land before a person could plausibly have seen the email.
  • Collapse into the delivery window. When a recipient's activity all happens around delivery and nothing follows, that's a scan, not a read.
  • Late rescans. The same scanner signature returning minutes or hours later, which otherwise looks like a real read.

Browser fingerprint

  • Stale signature fan-out. One identical, out-of-date browser version showing up at many unrelated companies at once.
  • Rotating devices. The same email clicked from Windows 10, then Windows 7, then Android within a minute. No person switches devices that fast.
  • Known scanner history. Browser signatures that have been machines almost every time they've appeared before.
  • Proxy signatures. Apple, Google, Yahoo, and AOL image proxies, each handled by its own rules.

Network

  • Shared network blocks. One block of addresses serving dozens of unrelated companies in the same minute.
  • Security vendor and cloud ranges. Traffic from gateway, hosting, and data center networks instead of homes and offices.
  • Session inheritance. When one event in a burst is clearly a scanner, its siblings from the same network share that verdict.
  • Company-wide patterns. When every recipient at one company engages within the same few seconds, it's that company's gateway.

Link behavior

  • Every link clicked. A person clicks one or two links. A scanner clicks all of them.
  • Unsubscribe in the first minutes. Unsubscribe links clicked during the delivery scan.
  • Click bursts. Many clicks on one email inside a few seconds.
  • Clicks with no open. Weak evidence on its own, since readers with images turned off do this. Counts only alongside other signs.

Provider signals

  • SendGrid's own machine flag. Used when SendGrid marks an Apple privacy open, then checked against everything else.
  • Gmail prefetch timing. Gmail proxy opens right at delivery lean machine. Gmail proxy opens later lean person.

How real readers stay real

Calling a person a machine is the worse mistake, because it hides a real lead. So HumanOpens leans toward people when evidence is thin.

  • Gmail readers count. A Gmail proxy open long after delivery is treated as a person reading, not a bot.
  • Fast readers count. Someone opening from a phone notification within a minute isn't flagged on timing alone.
  • Images-off readers count. A click with no recorded open needs other evidence before it's called a machine.
  • One clean human event wins. If any open or click on an email came from a person, that email counts as read by a person.
  • Unclear stays unclear. Events that could go either way are labeled that way in your report and CSV, instead of guessed.

Why this isn't a filter you set once

  • The fingerprints move

    Security vendors update browser versions, add networks, and change how many passes they make. A rule that worked in spring quietly stops working by fall. HumanOpens watches a live daily send to corporate inboxes and updates its checks when the patterns shift.

  • Single rules break

    "Ignore anything in the first 60 seconds" throws out real fast readers and misses late rescans. "Ignore Gmail proxy opens" throws out every Gmail reader. The answer comes from weighing many signals together.

  • The best clues are across companies

    A scanner looks normal one event at a time. It gives itself away when the same signature hits hundreds of unrelated companies in the same minute, which you only see by looking at everything at once.

  • It's tested against known answers

    Every change is scored against labeled traffic before it ships, with the rate of real people wrongly flagged tracked as closely as the scanners caught.

See it on your own data

Connect SendGrid for 7 days and get every send checked, with a contacts CSV of who really clicked and opened.

Start my free audit