Gmail 2026: what actually changed in the filter

Gmail does not publish changelogs for its spam filter. What senders get instead is a trail of documentation updates, a handful of Gmail engineering blog posts, and — loudest of all — a shift in their own Postmaster Tools graphs. Between October 2025 and February 2026 a lot moved. This article covers what, how to detect it, and what to change.

A short history of how the filter got here

Gmail's filter has always been a ML model fed by signals. The evolution, compressed:

• 2004–2012: Bayesian content filters, rule-based heuristics.
• 2013–2017: tabs (Primary, Promotions, Social, Updates, Forums) + reputation signals from SPF, DKIM.
• 2018–2023: TensorFlow-based models, engagement becomes dominant, DMARC expected (not required).
• Feb 2024: bulk-sender requirements announced with Yahoo. List-Unsubscribe, DMARC, spam complaint caps.
• Late 2025 – Q1 2026: enforcement of the 2024 requirements moves from "warn" to "reject". Engagement weighting and ARC validation tighten.

The 5 key 2026 changes

1. Engagement sampling window shortened

Before: Gmail weighted recipient engagement over a trailing 60–90 days. Now: the short-term window (7–14 days) carries more weight. A single bad week hits harder. A fixed campaign recovers faster. Senders who were coasting on historical engagement saw placement shift in December–January with no obvious cause.

2. DMARC reject propagation is now enforced end-to-end

Bulk senders who failed to publish DMARC after the 2024 deadline got warnings through 2025. In Q1 2026 Gmail began rejecting unauthenticated bulk mail at the envelope — not silently routing to spam, but issuing550 5.7.26. If your ESP dashboard shows bounces you can't explain, and your From-domain lacks a DMARC record, this is probably why.

3. DBEATI (domain-based engagement aggregation) increased

Gmail has aggregated engagement at the domain level for years, but the relative weight of domain-level vs IP-level reputation shifted further toward domain in 2026. Practical effect: rotating IPs no longer helps as much as it used to. A bad domain on a clean IP is still a bad domain.

4. ARC chain validation enforced for forwarders

Mailing lists, corporate forwarders and Google Groups-style redistribution used to break DKIM. ARC (Authenticated Received Chain) was the proposed fix; in 2026 Gmail began requiring ARC signatures from intermediate forwarders before trusting the original auth. Senders who rely on forwarding to hit subscribers may see sudden drops in engagement on a subset of recipients.

5. Bulk-sender compliance has teeth now

The 2024 rules — List-Unsubscribe with HTTPS POST, spam rate under 0.3%, DMARC alignment — were announced with a quiet enforcement period. That period ended. Exceed 0.3% complaint rate on Gmail now and the filter aggressively moves you to Spam across the board, not just for that campaign.

How to detect each change in Postmaster Tools

Google Postmaster Tools (postmaster.google.com) exposes most of what you need. If you don't have it set up, do that first. Then look for:

• Domain reputation dropping a tier (High → Medium or Medium → Low) without a volume spike — likely DBEATI weighting.
• Authenticated traffic % dropping — usually ARC chain break on forwarded mail.
• Spam rate edging toward 0.3% — the hard ceiling. Address it before you cross.
• Delivery errors with 550 5.7.26 — DMARC enforcement.

What to change if you are affected

1. Publish or tighten DMARC. Minimum v=DMARC1; p=quarantine; rua=... with alignment working. If you're at p=none, move to p=quarantine after two weeks of clean reports.
2. Clean the list before the next big send. Shortened engagement windows punish dormant-list blasts harder than ever.
3. Validate ARC on forwarders. If you send to mailing lists, confirm they add an ARC header. Google Groups does. Many legacy enterprise forwarders do not.
4. Segment by engagement. Send first to your top 30% engaged cohort. Let the short-window signal go positive before the broader send.
5. Run a placement test before and after each change — 20+ providers, free, unlimited on Inbox Check.

Warm-up implications

Warm-up tools that simulate engagement still work, but the shorter sampling window means warm-up signals fade faster. Plan to run warm-up continuously at low volume rather than in 30-day bursts. If you pause a domain for four weeks, expect to see reputation regression by the time you restart.

Cold-outreach implications

Cold outreach to Gmail got harder. Not impossible — personalization and tight targeting still win — but the margin for error shrank. Specific implications:

• Strict DMARC is now mandatory, not optional. p=none cold-outreach domains see rejects at scale.
• Daily volume caps of 30–50 per mailbox remain the realistic ceiling. Pushing to 100 per mailbox trips the short-window engagement signal.
• Reply rate matters more than open rate now — engagement weighting favours conversational signal over passive signal.
• Inbox rotation still works, but rotating across 20 weak domains is worse than running 3 strong ones. DBEATI punishes weak domains faster than before.

Frequently asked questions