What unique signals do Gmail track (engagement, recency, machine learning)?

You've probably noticed that two people on your list can receive the exact same email and end up with completely different experiences. One sees it in the Primary tab, the other finds it buried in Promotions. That's not random. Gmail builds an individual model for each of its users, and it uses that model to decide where your email belongs for that specific person.

Here's what Gmail is actually watching.

The signals that carry the most weight

Replies are the strongest positive signal Gmail tracks. When someone replies to your email, it tells Gmail's system that a real relationship exists between those two addresses. It's the closest thing to a trust handshake that email has. Starring an email or marking it as important also carries meaningful weight, though not as much as a reply.

Opens and clicks still matter, but Gmail is thoughtful about how it reads them. A click within the first few minutes of delivery signals genuine interest. A click three weeks later, after someone digs through their archive, lands much softer. Speed of engagement is a real factor. Gmail notices when a subscriber opens your email immediately versus letting it sit for days before acting.

Folder behavior is the one that catches senders off guard. Moving an email out of spam is a strong positive signal. Moving it into spam is an obvious negative one. But archiving without opening, or deleting without reading, those are quiet negative signals that pile up over time. Gmail tracks all of it.

How recency weighting works

Gmail doesn't treat a six-month-old engagement the same as one from last week. Recent behavior is weighted more heavily than historical behavior. So if a subscriber used to open every email but has gone quiet for three months, their current model reflects that change. You can't coast on a strong relationship that's gone cold. Gmail will notice before you do. (This is exactly why Gmail inbox placement can shift suddenly even when nothing in your setup has changed.)

Where machine learning fits in

Gmail uses machine learning to predict, at the moment of delivery, whether a given subscriber is likely to want a given message. It's not a static score. It's a live prediction layered on top of all the signals above, combined with the content of the email itself, the sender's global reputation, and patterns Gmail has observed across its entire user base.

This is why the same domain can land in Primary for one recipient and Promotions for another. It's not that your domain is categorized one way globally. It's that each user model produces a different prediction. Your relationship with each individual subscriber is genuinely tracked at an individual level.

What this means for your sending

Still the practical takeaway is that Gmail rewards real engagement, and penalizes its absence. Sending to subscribers who stopped engaging months ago doesn't just waste resources. It actively chips away at your placement for the rest of your list (this is where understanding the difference between per-user filtering and global sender reputation really matters). Keeping your list clean and prioritizing your most engaged subscribers isn't just good hygiene. It's the best thing you can do for Gmail placement.

Now if you want to understand which subscribers are still engaging and which have gone quiet, our SOS hotline is free and we're happy to walk through your list health with you ;)