Spam trigger words: mostly myth, partly real

Spam trigger words are mostly a myth. Modern filters at Gmail and Microsoft 365 score sender reputation, authentication and recipient engagement far more heavily than any list of forbidden vocabulary, so a single "free" or "guarantee" will not send a healthy account to spam. The part that is real: on a domain with thin reputation, content signals carry more weight, and copy that stacks several risk markers at once can tip the score.

Where did the trigger-word myth come from?

It was true once. Early-2000s content filters such as SpamAssassin worked largely from scored rules: phrases like "act now" or "100% free" each added points, and past a threshold the message was junked. Marketers reverse-engineered those rules, published word lists, and the lists have been recycled in blog posts ever since.

Filtering moved on. Gmail and Microsoft now run machine-learnt systems trained on enormous volumes of mail and, critically, on how recipients behave — what they open, reply to, delete unread, or report. The word lists did not move on, because they are convenient: rewriting an email costs an afternoon, while fixing infrastructure takes weeks. If you want the full infrastructure picture — domains, authentication, warm-up, verification, monitoring — start with the Cold Email Deliverability guide. This article covers the narrower question of whether your words are the problem. Usually they are not.

What do modern filters actually weigh?

In rough order of importance:

1. Sender reputation — the history of your domain and IP: complaint rates, spam-folder placements, prior engagement.
2. Authentication — SPF, DKIM and DMARC aligned and passing.
3. Engagement — replies and opens help; deletions without reading and spam reports hurt.
4. List quality — bounce rate is the fastest way to wreck a sender score, because every bounce tells the filter you are emailing addresses you do not know.
5. Volume behaviour — steady, human-scale sending versus sudden bursts.
6. Content — words, formatting, links. Last on the list.

A warmed mailbox scoring 91/100 on reputation can write "free consultation" and land in the inbox. A cold domain with no history can send flawless prose and still land in spam. That asymmetry is the whole story.

When do words still hurt you?

The mechanism runs like this. When your domain is new or lightly used, filters have little engagement history to score, so they lean harder on content. When the content then stacks several markers at once — capital letters in the subject, urgency phrasing, multiple links, a link shortener, image-heavy HTML, an attachment — each adds weight to the same side of the scale. When a few early recipients report the message as spam, that template's fingerprint becomes associated with complaints, and every later send of the same copy inherits the penalty. Then the spiral: lower placement, lower engagement, lower reputation, lower placement again.

So the honest answer is conditional. On a healthy domain, "guarantee" is harmless. On a fresh secondary domain with two weeks of history, a caps-heavy subject line with three tracked links is genuinely risky — not because of any single word, but because content is the only evidence the filter has.

What actually flags B2B cold email?

Patterns, not vocabulary:

• Link shorteners. Heavily abused by actual spammers, and scored accordingly.
• Open-tracking pixels. An invisible image loaded from a sending tool's domain is a recognisable fingerprint.
• Heavy HTML. Templates, banners, buttons. A genuine one-to-one business email is plain text; look like one.
• Attachments in a first email from a stranger.
• Identical bodies at scale. Thousands of byte-identical messages are trivially fingerprinted; light personalisation breaks the pattern.

Sector vocabulary is fine. Recruitment emails full of "opportunity", "placement" and "fee" deliver perfectly well when the infrastructure underneath is sound — outbound for recruiters covers that market's two-sided pipeline in detail.

So what should you change in your copy?

Write like one person emailing another. Plain text. Sentence-case subject, ideally under six words. No images, no attachments, at most one link — and preferably none in the first touch. Short paragraphs. A question, not a pitch, as the close.

This matters less for dodging filters than for earning replies, and replies are themselves a deliverability signal: at around a 4% positive-reply rate you are feeding the engagement side of the score with every send. Below 3%, fix the targeting and the offer — not the synonyms.

Where do words rank in the priority list?

Fifth of five. Authentication first, list verification second, volume discipline third, engagement fourth, content fifth. If your email lands in spam, the diagnosis almost always sits in the first four. Rewriting "free" as "complimentary" treats the symptom you can see instead of the mechanism that scores you.

---

Next step: the Growth System Audit — £450, seven days, credited against any build — maps where your growth system leaks and what to build first.