How Gmail decides inbox placement

Gmail does not make one decision about your email. It makes several, in order, and they use different inputs. Understanding which decision you are failing is the difference between fixing a problem and rewriting subject lines for a month.

The three numbers

Your ESP reports the first. Most tools claim to report the second. Almost nothing reports the third, and the third is what determines whether a human replies.

Decision one: acceptance

This stage is largely mechanical and the requirements are published. SPF and DKIM must pass with DMARC alignment on the visible From domain. The sending IP needs valid forward-confirmed reverse DNS. TLS on the connection. Bulk commercial mail needs one-click unsubscribe (List-Unsubscribe plus List-Unsubscribe-Post), honoured promptly. Fail these and you get SMTP rejections — 550 and 554 responses that name the reason.

Failures here are the cheapest to fix because they are deterministic. Send a message to an external mailbox, read Authentication-Results, run dig -x on your sending IP, and you know.

Decision two: inbox or spam

Now it gets statistical. The inputs that matter most:

• User-reported spam rate. Google's threshold for bulk senders is 0.30% and the sane internal target is under 0.10%. This is the one number Gmail hands you directly.
• Bounce and invalid-recipient rate. Mail to addresses that do not exist is the clearest available evidence that your list was not collected from real people.
• Engagement history for your domain. Opens are weak, replies are strong, and a message a human drags out of Spam into the inbox is the strongest single positive signal you can generate.
• Volume consistency. A sender with a stable daily shape is predictable; a sender that jumps an order of magnitude looks like either a compromised account or a list purchase.
• Link and infrastructure reputation. Your links, your tracking domain, your shortener, your sending IP neighbourhood.

A cold sender starts with none of this history. That is the entire case for warm-up: it is not a trick to inflate a metric, it is the production of real engagement signals for a domain that has none, at a volume Gmail finds unremarkable.

Decision three: tab and prominence

Once a message is inboxed, Gmail sorts it. Promotions and Updates are populated from message structure — bulk headers, HTML template complexity, image-to-text ratio, link count, marketing signature blocks — plus the absence of a relationship with the sender. Prior replies between two addresses pull strongly toward Primary.

The critical measurement trap: Promotions still carries the `INBOX` label. A message in Promotions is labelled INBOX and CATEGORY_PROMOTIONS. Every label-based placement check counts it as an inbox win, which is why reported Gmail placement is systematically optimistic. If a report gives you one Gmail number with no tab breakdown, it is measuring "not in Spam".

What you can actually see now

Postmaster Tools is the only first-party view, and it changed. v1 was retired — senders were redirected to v2 from 30 September 2025, and v1 was gone by the end of 2025. The domain and IP reputation dashboards are permanently gone. v2 reports spam rate, authentication pass rates, delivery errors and compliance status against the bulk-sender requirements, with a Deliverability analysis report added in June 2026.

So the reputation grade that half the internet still tells you to check does not exist. Build your monitoring on spam rate, delivery errors and compliance, and get placement from seed testing instead.

Where to intervene

1. Rejections at the SMTP layer → authentication, rDNS, blocklists. Deterministic, fix first.
2. Spam-folder placement → spam rate, list quality, volume shape, engagement history. Slow, and the only durable work.
3. Promotions instead of Primary → message structure and relationship. Fast to test, and only worth doing after (2) is healthy.

Working them out of order is the most common waste of effort in this subject: rewriting templates while a 0.6% spam rate is burning the domain.

What not to do

• Do not optimise for opens. Image proxying and prefetch make open data a poor signal, and Gmail treats replies and folder rescues as far more meaningful.
• Do not drive volume up to dilute a spam rate. It is a ratio, and a rising curve is itself a negative signal.
• Do not interpret a delivery rate as deliverability. Accepted mail includes spam-foldered mail.
• Do not send cold campaigns from your root domain. Use a subdomain so a bad campaign cannot damage transactional mail.
• Do not hunt for your domain reputation grade. Any tool still showing one is estimating it.
• Do not treat a single seed test as truth. Filtering varies per recipient; you need several seeds and a series over time.

How MailPilot helps

MailPilot addresses the two decisions you can influence. It warms Gmail and Google Workspace mailboxes over OAuth — and SMTP/IMAP senders including SendGrid, Amazon SES, Mailgun and Postmark — generating genuine replies and folder rescues at a ramp that stays unremarkable. Then it measures per-provider inbox placement with a seed-inbox network that reports Gmail's tabs separately, checks SPF, DKIM and DMARC, and tracks each mailbox's reputation daily.