Gmail's filtering models weigh replies, moves, stars, and deletes far more than the open rates senders obsess over. What the signals are and how to influence them.
Gmail decides where your mail lands using models trained on what its users do with it. Senders spend their energy optimizing open rates, a metric Gmail neither reports to you nor particularly trusts internally. Understanding which behaviours the filtering models actually observe changes how you segment, how you write, and how you decide who to stop mailing.
Placement is a per-recipient decision
Gmail's spam and categorization decisions combine global sender reputation with individual recipient history. A sender with a High domain reputation still lands in spam for the specific user who deleted the last twenty messages unread. The reverse also holds: a user who consistently rescues your mail from spam teaches Gmail to inbox you for them even while your broader reputation is weak.
This is why aggregate inbox placement rates from seed tests only approximate reality. Seeds have no behavioural history, so they measure the cold-start decision, not what your engaged subscribers experience. Both numbers are useful; they are just different numbers.
The signals that help you
Google has publicly confirmed the broad categories its models use, and deliverability practice fills in the weighting. The strongest positive action is a reply, because it is unambiguous evidence of a wanted correspondence. Moving a message from spam to inbox or from Promotions to Primary is nearly as strong, since it directly corrects a filter decision. Starring, forwarding, and adding the sender to contacts all register as affirmative interest.
Reading behaviour matters in subtler ways. Opening a message and dwelling on it is a weak positive; opening, scrolling, and clicking is stronger. These are observed client-side by Gmail itself, which is why the models have no need for your tracking pixel and are unaffected by whether it loads.
The signals that hurt you
The report-spam click is the loudest negative signal and feeds both the personal model and your domain-level complaint rate. Below it sits the delete without reading, individually weak but corrosive in aggregate: a subscriber who reflexively swipes away every issue is training Gmail to stop inboxing you for them. Sustained ignoring, where messages accumulate unopened, reads similarly over time.
Unsubscribing is not a negative reputation signal in itself. Gmail treats a clean unsubscribe as the system working as intended, which is one more reason to make leaving effortless. The subscriber you make hard to lose becomes the complainer you cannot afford.
How Gmail reads recipient behaviour
| Feature | Builds reputation | Erodes reputation |
|---|---|---|
| Explicit action | Reply, forward, star | Report as spam |
| Filter correction | Rescue from spam, move to Primary | Move to spam manually |
| Passive pattern | Open, scroll, click over time | Delete unread, sustained ignoring |
| List management | Add sender to contacts | Repeated filtering to trash by user rule |
Why your open rate is not the model's open rate
Sender-side open tracking counts image loads against a unique pixel URL. Proxy prefetching, image caching, corporate security scanners, and privacy features all fire that pixel without a human reading anything, while image blocking suppresses it for genuine readers. Gmail observes the actual client behaviour and does not depend on any of it.
The practical consequence: treat your open rate as a trend indicator with known noise, not as the quantity Gmail is judging. A stable open rate with declining click and reply activity can still accompany a placement slide, because the signals Gmail weighs were deteriorating while your pixel kept firing.
Turning this into an operating policy
Engagement-first sending rules
- 1
Define engagement windows per stream
For example: active means opened or clicked within 90 days for a weekly newsletter, within 180 for a monthly digest. Write the definition down and enforce it in the segment logic.
- 2
Reduce frequency before removing
Move fading subscribers to a lower cadence first. Fewer, more relevant sends often revives the pattern Gmail wants to see.
- 3
Run one reactivation attempt, then stop
A single win-back message with an explicit choice beats months of mailing into silence. Non-responders get suppressed, not archived for next quarter's blast.
- 4
Watch per-provider engagement separately
Segment metrics by mailbox provider. A Gmail-specific engagement slump is an early placement warning that blended averages will hide.
Frequently Asked Questions
Does landing in the Promotions tab hurt reputation?
Can I recover a recipient who stopped engaging?
Do these signals exist at other providers?
How fast do personal signals change placement?
Gmail is measuring whether individual humans demonstrably want your mail. Every list decision that increases the share of recipients who reply, click, and rescue, and decreases the share who delete unread, moves the models in your favour. Send to people who act like they want it, and the algorithm has nothing to hold against you.
Key Takeaways
- Gmail's placement decisions are personalized: the same campaign can inbox for one user and folder for another
- Strong positive signals are replies, rescues from spam, moves to primary, stars, and adding the sender to contacts
- Strong negative signals are spam reports, deletes without reading, and sustained ignoring
- Open tracking pixels measure image loads, not attention; Gmail's models do not need your pixel
- Engagement-based sunsetting is the most direct lever a sender has over these signals
Related articles
2025 in Email: The Year Authentication Became Mandatory Everywhere
Microsoft joined the mandate, Gmail moved to hard rejection, and DMARCbis reached the finish line. What 2025 changed for senders and what it sets up for 2026.
Building a Deliverability Monitoring Stack
Postmaster Tools, SNDS, FBLs, DMARC reports, TLS-RPT, bounce logs, engagement data: the full observability stack, what each layer catches, and the alerts worth paging on.
Q4 Peak Sending: Ramping Volume Without Triggering Filters
Between Black Friday and year end, send volumes triple while filters tighten. How to ramp into peak season so November's volume looks like growth, not an incident.