Research8 min read

A field guide to the internet's long tail: 36,377 unmatched MX hostnames

Every classifier has a remainder. Ours is 36,377 unique MX hostnames that match none of our 310+ patterns, spread across 81,514 domains — one in eight of the top million. This is what actually lives there.

Most writing about email infrastructure describes perhaps a dozen companies. That is fair as far as it goes — the concentration is real — but it quietly ignores a population that our daily email infrastructure report measures every night and most surveys throw away. As of the 2026-07-05 snapshot, our classifier — 310+ regex patterns applied to the primary MX of every domain in the Tranco top-1M — leaves 36,377 unique MX hostnames unmatched. Together they put 81,514 domains, 12.45% of everything that publishes an MX record, into the Unknown/Generic bucket. Add the 147,712 domains (22.55%) we classify as self-hosted, and roughly a third of the top million runs on mail infrastructure that no deliverability blog has ever named.

A taxonomy of the unclassifiable

"Unknown" does not mean random. Read a few thousand of these hostnames — the top of the distribution is published as long_tail_top1000.csv — and clear families emerge.

1. Regional hosters below the pattern radar

The largest family: web-hosting and telecom providers serving one country or one language market. Each runs mail for hundreds or thousands of customer domains on hostnames only its own customers would recognise. Individually, none justifies a classifier pattern; collectively, they are the backbone of the tail. This is also why the tail has a geography — the further you get from the English-speaking SaaS market, the larger the share of domains it holds.

2. Security appliances with per-company names

On-premise and boutique-cloud filtering gateways often give every customer a bespoke MX hostname — frequently under the customer's own domain or a per-tenant label of the vendor's. The result is infrastructure that is vendored but not fingerprintable: hundreds of hostnames, one product, no shared string to match. Some government domains sit in exactly this family, riding appliance MX hosts that give no outward hint of the vendor behind them.

3. True self-builds

The classic mail.example.com pointing at a box the organisation genuinely runs: a Postfix or Exim on a VPS, an old Exchange behind a static IP, a university department server that predates the cloud. These overlap with our explicit self-hosted class; the ones landing in Unknown are those whose naming gives no generic signal at all.

4. The broken and the abandoned

The tail's basement is configuration that cannot work, at scale that deserves its own numbers:

  • 4,925 domains publish an MX record with an empty target — a record that says "mail goes to" and then names nothing.
  • 502 domains point their MX at localhost, instructing every sender on Earth to deliver the mail to itself.
  • 157 domains point at a literal ~, and 253 at a bare mail with no dot — a relative hostname that resolves nowhere, or worse, somewhere unintended.

Most of these are attempts to say "this domain does not accept mail" by someone who did not know the standard way to say it: Null MX, 0 ., per RFC 7505. We cover that pattern and its impostors in our Null MX analysis, and the genuinely strange specimens — hostnames past 70 characters, typo-chains that concatenate two providers' domains into one impossible FQDN — get the tour they deserve in the weirdest MX records in the top million.

Why the tail matters if you measure email

For a measurement project, the tail is the honesty test. It is easy to publish clean market-share charts if you silently drop whatever your patterns miss; the 12.45% figure exists because we count our remainder instead. Two methodological notes follow. First, the tail is a lower bound on diversity: our classification reads the primary MX (lowest preference), and CNAME-fronted MX targets we do not chase can hide a known provider behind an unknown name. Second, the tail is where classifier drift shows up first — a regional hoster growing fast appears here years before it earns a pattern. The full pipeline, including where it breaks, is documented in our methodology write-up.

Why it matters if you send email

The long tail has no playbook

Deliverability advice is written for Gmail and Microsoft 365. None of it transfers to 36,377 hostnames running who-knows-what filtering with who-knows-what strictness — there is no postmaster page, no feedback loop, no documented thresholds.

Practically, sending into the tail means: conservative per-domain volumes, clean plain-text-friendly content, tight bounce monitoring, and no assumption that a technique validated on Gmail generalises. One in eight prospect domains lives here. The operational side of that problem — how to run outreach when the receiving stack is unidentifiable — is covered in our outreach guide to the unknown-MX long tail. The dataset side updates nightly on the long-tail page, CSV included, if you want to see which unmatched hostnames carry the most domains this week.

FAQ

What exactly counts as an 'unmatched' MX hostname?

A hostname that appears as the primary MX (lowest preference) of at least one Tranco top-1M domain and matches none of the 310+ regex patterns in our classifier. There were 36,377 such unique hostnames in the 2026-07-05 snapshot, covering 81,514 domains.

Is Unknown the same as self-hosted?

No. Self-hosted (147,712 domains, 22.55%) is an explicit classification for domains whose MX signals genuinely own-run infrastructure. Unknown/Generic (81,514 domains, 12.45%) is the remainder our patterns cannot place at all — a mix of regional hosters, appliance gateways, self-builds with uninformative names, and broken records.

Why not just add patterns until the tail disappears?

We add patterns continuously, but the tail is structural: thousands of providers each serving a handful of domains, appliances with per-customer hostnames, and one-off self-builds. Chasing every hostname would trade a quantified remainder for a false sense of completeness — and dictionary growth itself distorts trend lines, which is a documented caveat of our methodology.

Where can I get the raw long-tail data?

The long-tail page of the daily report ranks unmatched hostnames by the number of domains they serve, and long_tail_top1000.csv is downloadable for offline analysis. Both regenerate nightly from the OpenINTEL-based scan.
Related reading
Found this useful? Share it
AB
About the author
Artem Berezin
B2B Deliverability Specialist

B2B deliverability specialist with 5+ years of hands-on outreach experience. Built campaigns reaching 90,000+ inboxes across 20+ countries — and fixed the deliverability problems that came with that scale.

Check your deliverability across 20+ providers

Gmail, Outlook, Yahoo, Mail.ru, Yandex, GMX, ProtonMail and more. Real inbox screenshots, SPF/DKIM/DMARC, spam engine verdicts. Free, no signup.

Run Free Test →

Unlimited tests · 20+ seed mailboxes · Live results · No account required