Email deliverability in 2025 — what changed, what still works

2025-05-14 · operations · 6 min read · Pierre Richard

The deliverability rules quietly tightened twice in the last year. If you set up your sending infrastructure in 2023 and haven't revisited it, there's a real chance your inbox placement has been drifting down without you noticing — Postmaster reports show it lagged by 2-3 weeks. Here's what shifted, and the warmup pattern we now use on every new domain to keep placement above 99%.

What changed

Google: bulk-sender requirements (Feb 2024 → enforced 2025)

Google's "email sender guidelines" went from suggested to enforced through 2024 for any domain sending >5,000 messages/day to gmail.com addresses. The three big asks:

SPF, DKIM, and DMARC must all pass. No more skipping DMARC; Google now treats absence of a policy as a soft fail. We default to p=quarantine; pct=100 on new domains.
One-click unsubscribe via List-Unsubscribe-Post header. RFC 8058. Has to actually work — sending feedback that doesn't unsubscribe will tank reputation.
Spam complaint rate must stay <0.3% (target <0.1%). Postmaster Tools show the rolling rate. Once you cross 0.3% sustained, placement falls off a cliff and recovery takes 3–6 weeks.

The third one is the operational gotcha. A 0.3% complaint rate sounds easy until you do the arithmetic on a rough list — one or two angry recipients per thousand and you're toast.

Microsoft: tenant-level reputation scoring (Q4 2024)

The under-reported change. Microsoft moved from per-IP to per-tenant reputation scoring on outlook.com / hotmail.com / live.com inboxes. If your sending domain shares any infrastructure (shared IPs on a transactional ESP, in particular) with a noisy neighbour, your placement degrades. Two implications:

1. Dedicated IPs matter more than they did. On AWS SES, the dedicated-ip-pool config is no longer optional once you're past ~50K/mo to Microsoft inboxes. 2. Subdomain isolation matters more than it did. Marketing should not share from: infrastructure with transactional. Period. We use mail.parisai.click for marketing and transactional.parisai.click for everything else, with separate DKIM keys and separate IP pools.

Yahoo / AOL: aligned with Google, mostly

Yahoo (which still includes AOL) adopted essentially the same bulk-sender requirements in early 2024. SPF/DKIM/DMARC, list-unsubscribe header, complaint-rate cap. Differences are minor enough to not worth covering separately.

The warmup pattern that still works in 2025

The basics haven't changed: gradually ramp send volume on a new domain to give receivers time to build a positive reputation profile. What did change is the ramp schedule and the kind of mail you send during it.

Our 30-day warmup for a new sending domain:

Day  1 :    50 mails to known-engaged addresses
Day  2 :   100
Day  3 :   200
Day  4 :   400
Day  5 :   800
Day  6 :  1,200
Day  7 :  1,600  → first weekend, hold flat
Day  8 :  2,000
Day  9 :  3,000
Day 10 :  4,000
Day 11 :  5,000
Day 12 :  6,000
Day 13 :  7,000
Day 14 :  8,000  → second weekend, hold flat
Days 15–30 : ramp to target steady-state, +20% per business day

Two non-obvious things in here:

1. The first 7 days must be to known-engaged addresses

"Known-engaged" means recipients you have evidence will open and respond to your mail. For us, that's our team mailboxes, design partners who've replied in the last 30 days, and a small cohort of engaged beta users we've explicitly told to expect warmup mails. This is not the warmup-tool seed-list pattern (those don't help anymore — Google detects the inbox-bot pattern). It's actual humans who'll actually engage.

If you don't have known-engaged addresses, the warmup will fail. Build the engagement layer first, then warm up.

2. The "weekend hold" reduces the ramp slope

Friday night sending, Saturday silence, Monday resume — but at the same volume you were at Friday, not at Friday × 1.2. Receivers are more lenient about a flat curve than a step function over the weekend. Costs you a day, saves you a placement drop.

What still works

The non-changing fundamentals worth re-stating because people forget them:

DMARC report monitoring. Spin up Valimail or DMARC Digests on _dmarc aggregate reports. You will catch a domain spoofing your from: within 48 hours of it starting; without monitoring, you find out when placement craters six weeks later.
Strict from-name pattern. Pierre Richard reads as a person; noreply@parisai.click reads as a list. Personal-name from: lifts placement on Gmail by 2-4% in our A/B logs.
Plain-text alternative. Always send a text/plain MIME alternative. Some receivers still penalise HTML-only mail. Costs you nothing.
HTML weight under 100KB. Heavy templates with embedded fonts get scored down. Our cold outreach is intentionally text-mostly with a single signature image.

What stopped working

Three patterns that worked in 2022 and don't anymore:

Buying "warmed" sending domains from marketplaces. They don't carry reputation across an ownership change anymore. The receivers track domain-DKIM-key-signing patterns and a key rotation triggers a fresh trust window.
Inbox-rotation tools (Mailshake-pattern multi-mailbox warmups). Google flags the synthetic engagement pattern. Don't.
Cold-list scraping. Even with perfect deliverability tech, sending to a list with >2% bounce rate burns reputation faster than warmup builds it. Validate before sending. We use BriteVerify; alternatives exist.

How we measure success

The single number we watch is Postmaster's "User-reported spam rate" on a 7-day rolling window. Anything below 0.1% means we're operating safely. 0.1–0.2% is yellow — investigate which campaign or which list segment is causing it. Above 0.2% we pause new sends from that domain and run an audit.

The supporting numbers (delivery errors, IP reputation, domain reputation) all matter, but they're lagging indicators. The complaint rate is what predicts placement first.

TL;DR

Most of the deliverability "secrets" still come down to: send wanted mail, send it from properly-authenticated infrastructure, ramp slowly, and monitor the right metric. The 2024-2025 changes aren't paradigm shifts; they're tighter enforcement of rules that were already there. The teams that get hit are mostly the ones who treated those rules as suggestions.

Want this on your prospect list? pierre@parisai.click.