Open rates are a number you pay for and get nothing back: should you turn off open tracking in cold email?
Yes. Turn open tracking off. Open rate is a number you pay for in deliverability and get nothing back for.
On a client weekly on May 6, the question came up the way it always does. A founder running investor campaigns wanted to know whether we should switch open tracking back on so she could see who was reading. We kept it off. Same answer I give everyone, and I'll walk through why.
Opens are a vanity metric you pay for
Open tracking works by putting a tiny invisible image in your email. When the recipient's mail client loads that image, your tool logs an open. That's the whole mechanism.
Two problems. First, it's not reliable. An open only means an image loaded. That can happen without a human reading anything, and it never happens for the people whose mail client blocks images. You end up making decisions on a number that's partly noise.
Second, and this is the one that costs you money: that image is a signal to the inbox provider. A plain text email from a person doesn't usually carry a tracking pixel. A mass email does. You're adding the fingerprint of a marketing blast to an email you want to look like a one-to-one note.
I saw a blunt version of this argument on X in April: open rates are a lie, and the only cold email number worth managing is positive replies. I'd go further. Open rates aren't only useless, they make the thing you actually care about worse.
What Gmail and the vendors say about tracking pixels
You don't have to take my word for it. The sending tools themselves have told their users to turn it off.
SmartReach's help documentation says Gmail flags emails that contain a tracking pixel, and that Gmail and Google Workspace users can see a warning banner when open tracking is on. Their recommendation is direct: to avoid being flagged, disable open tracking, and measure engagement through replies and, where it applies, clicks.
QuickMail went for a middle ground. They added a setting that keeps open tracking on overall but automatically turns it off for emails sent to Gmail or other providers you choose, so those recipients don't see the "images in this message are hidden" banner. Opens from those providers just aren't counted.
When the vendors who built open tracking start shipping ways to switch it off, that tells you where the number stands.
Our rule: replies and booked calls, nothing else
We judge every campaign on two numbers. Did people reply? Did they book? Everything else is decoration.
Replies tell you the list, the copy and the deliverability all worked at once, because a human read it and cared enough to type. Booked calls tell you the reply was real interest, not "please remove me." If a campaign has healthy replies, I don't care what the open rate would've been. If it has no replies, a high open rate wouldn't save it, it would just make you feel better while you burned a list.
Even replies need sorting before they count. A reply that says "interested, send times" is a win. One that says "talk to my partner, she handles this" is a referral and often better than a win. "Not now, try me in Q3" goes into a follow-up list with a date. "Remove me" gets removed, the same day, and counts for nothing. When I say we judge on replies, I mean the first two kinds. A campaign that gets lots of replies but all of them are removals is telling you the list is wrong.
The one thing open rate was ever good for was diagnosing spam placement. There are better tools for that: seed tests, inbox placement checks, watching bounce and reply trends per sending domain. Use those.
One inbox, nothing else
The other half of that May 6 call was about organization. The client was bouncing between the unified inbox, the CRM view and the campaign screens in her sending tool, and replies were slipping. The answer on the call: "Just unibox 100 unibox, nothing else. CRM, no campaigns, just unibox."
One place where every reply lands. One place your team checks. If a reply can live in three views, it'll get missed in all three.
We also found a nasty little bug in that workflow on the same call. Unanswered replies carry a blue marker so you know they need a response. But when a teammate clicked into a thread, the marker disappeared, even if nobody actually replied. So one person peeks, the flag vanishes, and the reply looks handled to everyone else. We also saw the inbox repeat emails after long scrolling sessions.
The fix is boring and works. The manual check on that call was simple: look for replies that start with a greeting to the client by name, because those are humans answering, not your own outbound. Whoever opens a thread owns it until it's answered. Don't trust a marker to mean "handled."
Where AI reply suggestions still need a human
The tool suggested replies with AI, and they were useful drafts. They weren't perfect, and the call logged it plainly: AI suggestions need human oversight.
For investor campaigns especially, I want a person reading every suggested reply before it goes. An AI draft that's 90 percent right can still say the wrong number, promise the wrong next step or answer a question nobody asked. The investor won't tell you it felt off. They'll just stop replying.
So the stack is: no open tracking, one inbox, replies and booked calls as the only scoreboard, and a human on every send that matters.
If you want to see the full system we run behind the inbox, here's how we work.
Frequently asked questions
Does open tracking hurt cold email deliverability?
It can. Open tracking adds a tracking pixel, and SmartReach's documentation says Gmail flags emails that contain one. Turning it off removes a marketing fingerprint from emails you want to look personal.
What should I measure instead of open rate?
Replies and booked calls. For deliverability problems, use inbox placement checks and watch bounce and reply trends per sending domain instead of opens.
How do I make sure no cold email reply gets missed?
Work every reply from one unified inbox, and don't trust an unread marker to mean a reply was handled. Whoever opens a thread owns it until it's answered.