Cold Email Opens Are High but Replies Are Zero, Bots, or Bad Copy?

Where the Phantom Opens Come From

• •Security gateways and filters: Microsoft Defender, Proofpoint, Mimecast and friends fetch every image and follow every link in inbound mail as part of scanning, before the recipient ever sees it, and regardless of whether it's then delivered to the inbox or junk. On protected domains this fires for every single email, which is how a campaign to gateway-heavy enterprises "achieves" 90% opens overnight. (Spot gateway-protected domains in advance with the ESP Segmenter, it reads MX records and flags them.)
• •Apple Mail Privacy Protection: since 2021, Apple Mail pre-fetches remote images through a proxy for all mail, marking essentially everything "opened" for any recipient using Apple Mail, a large slice of any B2B list's executives reading on iPhones.
• •Preview panes and prefetchers: some clients load images when a message merely scrolls past in the list. An "open" was never a promise that anyone read anything.

The Three-Step Diagnostic, in Order

• •Step 1, Autopsy the open timestamps. Export or eyeball per-recipient open times against send times. Machine signature: opens within ~60 seconds of delivery, uniform across a whole company domain. Human signature: spread over hours/days, business-hours clustering, repeat opens. If 60% of your "opens" are in the machine bucket, your real human open rate might be 15%, a completely different problem statement.
• •Step 2, Test placement directly. Send your exact campaign email to seed accounts you control at Gmail, Google Workspace, Outlook.com, and Microsoft 365. Inbox or junk? This answers what opens never can: whether humans are even being shown the email. Junk placement at one provider with fine placement elsewhere is the classic pattern, usually Microsoft, and the fixes live in the Outlook playbook. Verify your authentication is passing while you're at it with the Deliverability Checker.
• •Step 3, Only now, judge the message. Placement confirmed, human opens confirmed, still silent? That's a genuine copy or targeting problem: the wrong people, a first line that reads like a mail merge, no specific reason to reply, or an ask too heavy for a first touch. Rewrite against the response-focused framework and change one variable at a time.

The Deeper Fix: Stop Steering by Opens Entirely

This whole diagnostic exists because opens are a corrupted instrument. Our fleet runs with tracking off, no pixel, no rewritten links, and steers on reply rate, positive-reply rate, bounce rate, and weekly seed placement tests, none of which a scanner can fake. The full case (including what the pixel itself costs you at Microsoft) is in the open-tracking piece, and if you're keeping tracking during testing, at minimum run it on a custom tracking domain. A campaign judged on replies can't be fooled by robots politely opening everything you send.