State of Cold Email Deliverability 2026

Executive summary

The headline of 2026 is that cold email deliverability got measurably harder, and the reason it got harder is not the one the industry talks about. Authentication compliance — SPF, DKIM, DMARC — is near-universal among the senders we measured. What changed is what comes after authentication: the content-side fingerprinting systems at the major mailbox providers became substantially more aggressive between January and March of this year, and they punish senders whose campaign content does not resemble their warmup content.

The full report runs through seven findings. The short version:

1. 01Median inbox placement on real campaign content dropped 11 points YoY, from 74% in 2025 to 63% in 2026 — the steepest single-year decline NeverSpam has measured.
2. 02Senders using template-based warmup outperformed sender-only warmup by 28 percentage points (median inbox placement, n=14,392 campaigns).
3. 03Gmail’s content-fingerprinting weight increased meaningfully after the March 2026 algorithm changes, particularly for senders whose warmup content was generic.
4. 04Yahoo inbox placement is now harder than Gmail for cold email senders (39% versus 57% median). Outlook M365 is harder still in B2B contexts.
5. 051 in 4 senders with “perfect” DKIM/SPF/DMARC still landed in spam due to content fingerprinting alone. Authentication is necessary but no longer sufficient.

The implication is uncomfortable for a deliverability industry that has spent five years selling authentication audits and reputation dashboards: those signals still matter, but they no longer predict whether your campaign reaches the inbox. The variable that does predict placement — content similarity between warmup and campaign — is one most warmup vendors do not even measure.

Methodology — short version

We want to be honest about what this data is and is not, because most published deliverability statistics fail this test. The full version of the methodology is at the bottom of this report; the short version is here.

All numbers in this report come with implied 95% confidence intervals; medians are used in preference to means because the underlying distributions are heavily right-skewed. Sample sizes are shown alongside each finding. Where we extrapolate beyond what the data supports, we say so.

For data requests, methodology questions, or peer review, research@neverspam.com.

Finding 1 — The fingerprinting tax

Central thesis · 14,392 campaigns · 6 months

The single most important finding of this report is that sender reputation alone does not predict campaign placement in 2026. The correlation between a sender’s warmup score and the inbox-placement rate of their actual cold campaign content has collapsed.

To measure this we paired the warmup-side reputation score (NeverSpam internal, normalized 0–100) with the inbox-placement rate of the same sender’s real campaign content during the following seven days. We did this for every sender in our dataset who ran at least one cold outbound campaign during the research window. The result:

In plain language: two senders, both at warmup-score 90+, sending the same week, can have a 28 percentage-point gap in inbox placement. The variable that explains the gap is whether their warmup content matches their campaign content. Senders whose warmup probes used templates that resembled their real outreach landed in the inbox at a median rate of 81%. Senders whose warmup probes used generic “warmup-style” content landed at a median rate of 53%.

The pattern is symmetrical across every provider we measured. Sender reputation as expressed by warmup score predicts whether providers will even open the door to your mail — but the second gating function, content fingerprinting, decides whether your campaign actually gets through. Read our companion piece on why cold emails go to spam for the mechanics of that second filter.

Why this matters: a warmup vendor that reports your “score” without telling you anything about campaign-side content placement is selling you the speedometer of a car with no fuel gauge. The score climbs even as you run dry. For the full mechanics of how this is fixed, see our guide to template-based warmup.

Finding 2 — Provider-specific placement rates

By provider · cold email content · 6-month median

For years the industry has treated Gmail as the canonical hard target — the provider whose filters define the state of the art. As of early 2026, that is no longer accurate. Yahoo and Outlook M365 are both harder than Gmail for cold senders. iCloud is the friendliest of the four majors, partly because its filtering historically lags but more importantly because the iCloud user base receives significantly less B2B cold email and therefore has less negative-feedback training data.

Two observations are worth pulling out. First, the absence of a Promotions tab on Outlook and iCloud has the counterintuitive effect of making those providers binary: you are in the inbox or you are in spam, with no soft-landing area. Senders accustomed to using Promotions as a “close enough” outcome on Gmail face a harder yes/no everywhere else.

Second, Outlook M365 is where the steepest year-over-year decline happened — placement rates fell 14 points compared with our 2025 measurement. Microsoft’s shift toward neural filtering models for tenant-level mail flow has produced visible effects, and the senders most affected are B2B-to-B2B prospecting (i.e. exactly the cold email use case). For configuration-level help, see our guides on Microsoft 365 warmup and our email deliverability guide.

Finding 3 — The content patterns that get filtered

A/B paired sends · n=88,213 probes · Gmail + Outlook + Yahoo

We ran paired sends — identical sender, identical seed mailbox set, identical day-of-week and time-of-day — varying one content variable at a time. Every delta below is the median change in inbox-placement percentage versus the plain-text baseline.

Three patterns deserve commentary. First, link redirectors are catastrophic. The 22-point penalty for any bit.ly / lnkd.in / tinyurl link is the single largest content effect we measured. There is no situation in cold email where a redirector is worth that penalty; use a subdomain of your own root domain for tracking.

Second, the “unsubscribe link in cold outreach” result is initially surprising and ultimately not. Including a real unsubscribe link in cold email gives recipients an action that is not “mark as spam.” The substitution effect is large: a recipient who would otherwise have hit the spam button unsubscribes instead, and the negative feedback loop that would have damaged your sender reputation never gets created. Providers also infer from the presence of a working unsubscribe link that you are a sender who respects recipient choice. The +9-point effect is, in our reading, the most actionable single finding in this report.

Third, HTML wrappers and emoji in subject lines are still penalized in 2026 at roughly the same magnitude they were in 2022. The industry mostly stopped using these things, but new senders regularly rediscover them and pay the price. Read our cold email deliverability checklist for the full pre-send audit list.

Finding 4 — Warmup volume and ramp rates

Median daily sends · by vertical · among top-quartile placers

Senders who landed in the top quartile of inbox placement followed remarkably different ramp curves depending on what they were sending. We segmented by vertical and measured the median daily send volume at four checkpoints: day 1, day 7, day 14, and day 30 of a new sending identity.

Two observations. First, recruiting tolerates the most aggressive ramp by a wide margin — by day 30 a high-performing recruiting sender is doing nearly 100 sends/day from a single identity. Healthcare and finance, both regulated verticals, tolerate roughly a third of that.

Second, every successful curve in our data is sub-linear. None of the top-quartile senders doubled their day-14 volume by day 30. The pattern is a fast initial ramp tapering toward steady-state, not exponential. Senders who try to compress the curve fail. For prescriptive guidance, see our email warmup guide for 2026.

Finding 5 — When senders fail

Failure-mode taxonomy · n=3,118 senders below 35% inbox placement

We pulled every sender in our dataset whose cold-campaign inbox placement was below 35% during the research window — a group we informally call “the spam-bucket cohort” — and categorized the dominant cause of each failure. One sender, one cause; ties broken by largest measurable effect.

Content fingerprinting is, by a wide margin, the leading cause of failure. Roughly a third of struggling senders are filtered for reasons that have nothing to do with their authentication posture, domain age, volume curve, or even list quality. They are filtered because their content looks, to the provider’s ML stack, like email patterns the provider has previously decided are unwanted.

The second-largest bucket — sender reputation drift — is what every warmup product on the market claims to address. The third — authentication misconfiguration — is what every deliverability audit tool checks. Combined, those two represent 36% of failures. Content fingerprinting alone is 34%. The current deliverability tooling market is, in aggregate, focused on the second-most common failure mode while the largest one is barely measured.

Use our SPF checker and DMARC checker to rule out authentication issues quickly — they will eliminate the 14% bucket above in roughly twenty minutes — and then move to the content side. Our email-headers analyzer will show you whether your spam path is reputation-driven or content-driven from the headers alone.

Finding 6 — Best-performing patterns

Top-decile placers (95%+ inbox) · n=1,247 senders

What does a sender with 95%+ inbox placement do differently? We took the top decile of our dataset — 1,247 senders — and looked at what they had in common. The pattern is unsurprisingly consistent and unsurprisingly boring.

The list above is not exotic. It is not even new. What is new is the magnitude of the gap between senders who do these things and senders who do not. In a 2024 dataset, doing all six of these things bought you maybe an 8-point placement advantage. In our 2026 data the same six behaviors buy you a 41-point advantage. The cost of cutting corners has gone up.

Finding 7 — Industry verticals

By outbound vertical · 6-month median across providers

Not all cold email is filtered equally. We segmented the dataset by the sending organization’s primary industry and measured inbox placement across providers.

Recruiting and SaaS lead. Finance and healthcare trail badly, almost certainly because providers apply a regulatory-vertical penalty to cold outreach in those categories — and because the downstream cost of a spam complaint in regulated industries is high enough that providers conservatively over-filter. E-commerce sits in the middle on inbox but has the largest Promotions share, reflecting providers’ tendency to categorize commerce-pattern email there regardless of send context.

The vertical effect is real, but it is dwarfed by the within-vertical variance. A top-decile finance sender (62% inbox) outperforms a bottom-quartile SaaS sender (39%) by a wide margin. Industry is a starting point, not a destiny.

Implications for cold email teams

We try to keep these grounded in the findings above. Each takeaway maps to a specific finding and carries a measurable expected effect.

1. 01Treat authentication (SPF, DKIM, DMARC) as a baseline floor, not a strategy — it stops you from being filtered for the wrong reason, but it will not save you from a content fingerprint.
2. 02Audit your campaign content against your warmup content. If they are not topically similar, your warmup score is lying to you about real placement.
3. 03Move every redirector link to a tracking subdomain on your own root domain. The 22-point penalty for third-party redirectors is the single highest-ROI fix.
4. 04For Yahoo and Outlook M365 sends, drop volume targets by 40% versus your Gmail playbook. The thresholds are not symmetric.
5. 05Rotate sending identities every 90 days for any vertical with a regulatory profile (finance, healthcare). Provider memory of a struggling sender is roughly two quarters.
6. 06Plain text outperforms HTML for cold outreach in every provider we measured. Reserve HTML for permission-based sends only.
7. 07Add a real unsubscribe link to cold email. Counterintuitive, but the signal that you respect recipient choice has measurably positive placement effects.
8. 08Stop using the “Re:” prefix on first-touch cold email. Gmail’s thread-fingerprint check catches it and the penalty is steep.
9. 09Build your day-1 through day-30 ramp around your worst provider, not your best. The bottleneck is what defines real-world deliverability.
10. 10If your warmup vendor cannot show you provider-segmented placement on real campaign content (not warmup content), you do not have a deliverability signal — you have a vanity metric.

The shorter version of the whole report: the deliverability moat in 2026 is content, not authentication. Teams that internalize this and operate on it will be measurably ahead of teams that are still buying audits of records they fixed three years ago.

Methodology — extended

The seed mailbox network

NeverSpam operates a seed network of 100,247 real, human-owned mailboxes as of April 30, 2026. These are not synthetic accounts. Every seed mailbox belongs to a customer, a partner, or a participant in our incentive-compensated seed program. Each mailbox is age-verified (median account age 3.4 years), has a non-zero history of normal personal-use email, and has been opted into the network via documented consent. Seed mailboxes that show signs of disuse, abandonment, or anomalous send patterns are removed from the active pool weekly.

The network skew toward English-language, North American, and European mailboxes is intentional — this is where our customers send. We report findings only for provider/region combinations where we have at least 1,000 mailboxes and 50,000 monthly probes.

Probe definition

A “probe” in this report is one email sent from one sending identity to one seed mailbox, observed at the recipient side for folder placement within 30 minutes. We measured 12.4M probes over the research window, split roughly 8.2M warmup probes (sender-to-seed within the warmup network) and 4.2M campaign probes (real cold campaign content where the sender opted into measurement, with PII stripped and content fingerprinted but not stored).

Placement detection

Folder placement is detected via IMAP on the recipient side. We classify into five buckets: Inbox, Spam (or Junk, the Microsoft term), Promotions (Gmail/Yahoo only — we do not synthesize a Promotions equivalent for providers that lack one), Other (Updates, Forums, etc. on Gmail), and Missing (delivered to the receiving server per SMTP confirmation but not located in any folder within 30 minutes — generally indicating tenant-level quarantine on enterprise accounts).

What we excluded

• Probes from senders below our minimum credibility threshold (account age < 14 days, fewer than 100 lifetime sends, or domain registered within the research window).
• Probes to seed mailboxes with documented filter-rule customizations (custom forwarding, inbox-rule routing, vacation auto-responders during the probe window).
• Probes where the seed mailbox was offline or unreachable for >5 minutes of the post-send observation window.
• All consumer-permission marketing email, all transactional email, all internal-team email. This report is strictly about cold outreach.
• Any probe matching internal NeverSpam testing patterns (i.e. our own QA traffic).

Authentication context

Across the senders in our dataset, authentication compliance is near-universal: 99.2% of campaign senders had a valid SPF record, 97.8% had passing DKIM, and 91.4% had a published DMARC policy. This is why authentication does not explain variation in placement in 2026: it has become a floor, not a differentiator. For provider-side documentation see Google Postmaster Tools, the Microsoft Sender Support portal, Spamhaus, and DMARC.org.

Statistical notes

Medians are used throughout in preference to means due to right-skewed distributions. Confidence intervals on all percentages reported are within ±1.4 percentage points at 95% CI given sample sizes, except where explicitly noted. Comparisons across years (e.g. the “11-point YoY drop” cited in the executive summary) are between equivalent measurement windows in 2025 and 2026 using the same seed-network methodology, with sample-size adjustments noted in the dataset.

Reproducibility & contact

We do not currently publish the raw probe data because it includes recipient-side seed mailbox identifiers. We do provide aggregated, anonymized subsets to academic researchers, journalists, and peer deliverability vendors on request. To request data, ask methodology questions, or report errors, write to research@neverspam.com.

About NeverSpam

NeverSpam is a template-based email warmup platform built for teams that send cold email and care whether it lands. Unlike traditional warmup tools that measure sender reputation in isolation, NeverSpam warms a sender on content that matches the actual outreach the team is sending — which, as Finding 1 of this report demonstrates, is the variable that actually predicts inbox placement in 2026. See how it works and our companion piece on cold email statistics for 2026.

We publish original research because the deliverability industry runs on borrowed anecdotes and we have the data to do better. If you want to be notified when the next report drops, watch /research — the State of Email Warmup 2026 report is scheduled for Q3.