What Is a Spam Filter?
TL;DR
A spam filter is the software a receiving mail provider uses to decide whether an incoming message reaches the inbox, the spam folder or nothing at all. Modern filters weigh authentication results, sender reputation, content signals and, most heavily, how recipients at that provider have engaged with the sender before.
What signals does a spam filter use?
Identity first: does SPF or DKIM pass, does the passing domain align with the visible From address, and does the sending IP have a matching PTR record. These are cheap to check and expensive to fake, so they act as a gate before anything else is considered.
Then reputation and behaviour: what this domain and IP have sent recently, how often recipients marked it as spam, and whether anyone replies. Content matters least of the three, though it still matters at the margins, because a link to a domain with a bad reputation is a reputation signal wearing content clothing.
Is there one spam filter or many?
Many, and they disagree. Google and Microsoft each run proprietary filters trained on their own traffic and publish requirements rather than rules. Smaller domains typically sit behind an on-premise or gateway filter such as SpamAssassin or Rspamd, or behind a security vendor, each with its own scoring.
That is why placement has to be read per provider. The same message can reach the inbox at one large provider, be filed as spam at another, and be rejected outright by a corporate gateway with a stricter policy than either.
Can you test a spam filter before sending?
Partly. A content scanner tells you what a rule-based filter such as SpamAssassin would score, and a seed mailbox tells you where a real provider actually filed the message. Both are worth running, and neither predicts the outcome for a specific recipient.
The gap is engagement history. Seed mailboxes have never replied to you, so they under-represent placement for a domain your recipients already correspond with and over-represent it for a cold list of strangers.
Frequently asked questions
Far less than the lists of them suggest. Content weight is small next to authentication and engagement, and a filter that penalised the word free would misfile most legitimate commerce. Formatting that mimics bulk mail, and links to low-reputation domains, carry more weight than vocabulary.