A while ago an email I was expecting from a company went straight to Spam. Gmail had put a banner on top of it, “This message was marked as spam because previous messages from this domain were marked as spam”, with the sender’s domain in place of “this domain”. I once built an SMTP server that sent millions of emails a month, so I did what I’d have done back then. I saved the headers, the part of an email that servers write for each other and you normally never see, and went looking for the reason. I didn’t find it.
The message left the company’s Google Workspace and landed in Gmail, so it never left Google. Google’s receiving server ran four checks on whether the message came from where it claimed and wrote the results in the Authentication-Results header. All four passed. SPF checks the sending server against a list the domain publishes. DKIM checks a signature against a key the domain publishes. DMARC is the domain’s policy for when those fail, and this one is p=reject, the strictest setting, which tells receivers to refuse anything that fails. ARC carries the results along when a message is forwarded. So authentication wasn’t it.
What the banner says #
Gmail has explained Spam decisions with a banner since March 2012. The current set, which Al Iverson collected in October 2025, has a first-person one, “You reported this message as spam from your inbox”. Mine was about the sender’s past, tied to the domain, in the passive voice. I read that as weak evidence that it wasn’t me. Weak, because Gmail can learn things on my side without me clicking anything. An older message from the same domain that got filtered into Spam and that I never saw would count.
Google doesn’t say how far this banner reaches. The Email Log Search definitions for Workspace admins do show that it keeps two signals apart. “The message recipient reported previous, similar messages as spam” is one classification. “The sender has a poor reputation” is another, with a note that “spam reports about the sender domain result in poor reputation for the domain”. From my mailbox I can’t see which one fired.
What I could check from outside #
What a recipient can check lives in DNS, the public directory where a domain publishes its records. Anyone can repeat this with the domain and the sending IP from their own headers.
dig +short TXT <domain>
dig +short TXT _dmarc.<domain>
dig +short TXT <selector>._domainkey.<domain>
dig +short MX <domain>
dig +short -x <sending-ip>
That’s the SPF list, the DMARC policy, the DKIM key under the selector named in the signature, the MX and the name the sending IP maps back to. All of it agreed with what the receiving server had written down. The only thing DNS adds is the key size. Google asks for at least 1024 bits and recommends 2048, and a short key alone doesn’t send mail to Spam.
Then the blocklists, lists of domains or IPs known for spam that mail servers query over DNS. An empty answer means not listed, and it’s also what you get when the lookup is broken. So each list publishes a test point that has to answer, and I queried that first every time.
| List | What it covers | Test point answered | Subject |
|---|---|---|---|
| Spamhaus DBL | domains | 127.0.1.2 | not listed |
| Spamhaus ZEN | IPs | 127.0.0.2, 127.0.0.4, 127.0.0.10 | not listed |
| SURBL multi | domains in links | 127.0.0.254 | not listed |
| URIBL multi | domains in links | 127.0.0.14 | not listed |
| SpamCop | IPs | 127.0.0.2 | not listed |
| Barracuda | IPs | 127.0.0.2 | not listed |
The test points mattered. Spamhaus answers 127.255.255.254 when the query comes through a large public resolver, and the other lists gave me nothing at all through 1.1.1.1, 8.8.8.8 and 9.9.9.9. Without the test point both look like “not listed”. The table came from my ISP’s resolver.
A product, a newsletter or a support tool can also sign with the same domain people write from and send from somewhere else entirely. Google’s Postmaster Tools says domain reputation “only displays messages sent from the exact domain used for DKIM and SPF authentication”, so everything signed with that domain lands in one score.
Three guesses the headers can’t separate #
My own history with the domain. Google’s help says that emails you send to Spam make future mail from the same sender more likely to be filtered. If something from that domain got filtered without me noticing, Gmail may have learned from it. Medium.
The domain’s reputation at Gmail, fed by everything else signed with it. Google writes that “user spam reports lower a domain’s reputation” and that “some legitimate messages may be marked as spam”. Against it, no public list carries the domain, and a sender that publishes p=reject is usually watching its mail. Low to medium.
The sending path. The Received lines show whether the message came in through Gmail’s API, which is how tools send on a person’s behalf. Google publishes nothing saying that mail is scored differently, and the banner talks about the sender’s past. Low.
What would settle it #
Three experiments, cheapest first, and all of them need the sender.
- The sender sends the same message to three Gmail accounts that have never received anything from the domain. Inbox in all three points at my history. Spam in all three points at the domain.
- The same person sends me a plain message from the Gmail web interface, no tool in the path. If the banner is still there, the sending path is out.
- The sender reads domain reputation and the user-reported spam rate for the month in Postmaster Tools. It covers consumer Gmail only, isn’t real time and can hide low-volume days.
On my side I can search Spam and Trash for older mail from the domain, and check my filters and blocked addresses, before the 30-day Spam purge takes it.
What this doesn’t cover #
I didn’t run any of the three experiments. Talos, FortiGuard, VirusTotal and Safe Browsing couldn’t be read from where I ran this, so “no listing” means six lists. I don’t know what else the domain sends, how much, or to whom. And I haven’t searched my own mailbox yet, which is the first thing I should do.
I still don’t know why that message went to Spam. Google doesn’t write down which signal this banner maps to, and from the receiving side I can’t find out.