'Spam trigger words' are mostly a myth. Here's what filters actually read.
Somewhere right now, a marketer is rewriting "free shipping" as "complimentary delivery" because a blog post told them the word "free" triggers spam filters.
That advice is roughly fifteen years out of date. Those "200 spam trigger words to avoid" listicles describe how filtering worked in the mid-2000s — and following them today costs you clear copy while buying you nothing.
Where the myth came from
Early spam filters really did work off word scores: naive keyword systems where "FREE!!!", "viagra," and "act now" each added points toward a spam verdict. In that era, word choice genuinely moved outcomes, and the folklore fossilized.
Modern filtering at Gmail-scale is machine learning over thousands of signals — dominated by who is sending (authentication, domain and IP reputation) and how recipients respond (opens, replies, deletes, spam reports, "not spam" rescues). Content is one input among many, evaluated in context, not against a forbidden-words list.
The proof is in your own inbox: legitimate brands land "free," "sale," "% off," and "limited time" in your Primary tab every single day. And the actual spam in your junk folder often contains none of the classic trigger words — it's junked because of who sent it and how people reacted, not vocabulary.
The kernel of truth
Word choice isn't entirely irrelevant — it's just that the risky pattern is imitating spam's shape, not using individual words:
- ALL CAPS SUBJECT LINES with multiple exclamation marks!!! — the combination, not any word in it
- Deception patterns: "RE:" on a first-touch email, "you've won," fake forwarding — these are scored, and they also violate anti-spam law in most jurisdictions
- Filter-evasion tricks:
F.R.E.E, zero-width characters, text hidden in images — actively worse than the words they're hiding, because evasion itself is the signal - A stack of clichés at once: one "limited time offer" is fine; a subject line that reads like a 2004 spam corpus is a pattern match
Notice what unites these: they're things spammers do that legitimate senders don't. That's the actual rule.
What content signals still matter
If you're auditing content for deliverability, spend your attention here instead:
- The domains you link to. Link reputation is real and heavily weighted. Audit every link — including shorteners, affiliate redirects, and that one partner domain you never checked.
- Text-to-image balance. Image-only emails are a classic evasion pattern. Keep real, readable text in every send.
- Honest headers. Subject matches content; sender name is who you are. Deception is both a filter signal and a legal problem.
- A visible unsubscribe. Its absence is a spam signal; its presence routes exits away from the spam button.
- Clean HTML. Broken markup and hidden elements read as sloppy at best, evasive at worst.
The bottom line
Write "free" if the thing is free. Filters gave up on word lists over a decade ago; they judge you on identity, audience response, and spam-shaped behavior. Clear copy that says what you mean isn't a deliverability risk — it's the opposite: it gets clicks and replies, and those are the signals that actually move the score.