← All articles

Spam Trigger Words in Cold Email: What to Avoid in 2026

October 12, 2026 · 9 min read · by the 0effort team

TL;DR: No mailbox provider publishes a spam-word list, and one word like "free" rarely sinks a cold email alone. Gmail and Outlook judge mostly on sender reputation, authentication and recipient behavior, so risky words matter only alongside hype formatting, link clutter, templated sameness and spam reports. Test phrases with a split send, not a blacklist.

Most deliverability checklists still ship a spreadsheet of four hundred "spam trigger words," color-coded, to be scrubbed from every template before launch. We understand the impulse. A list feels like control, and deliverability otherwise feels like weather. We think it's the wrong main tool for 2026, and below is a test you can run in an afternoon instead.

Do spam trigger words still matter in 2026?

Less than the checklists suggest, and never in isolation. The folklore has a real origin. Open-source filters like SpamAssassin work partly by adding up rule scores: a hit for an all-caps subject here, an easy-money phrase there, and past a threshold the message is tagged. Word lists were a fair way to game that arithmetic.

What changed is who is deciding. Gmail, Outlook and Yahoo publish no banned-word list, and none has said a single keyword gets mail blocked. Their filtering is largely machine-learned, trained on enormous volumes of what recipients actually do with mail. When the model learns from behavior, vocabulary becomes one weak feature among many, and it carries weight only in company.

Our position: when a trigger word turns up, reread the sentence around it and judge the sentence.

How do Gmail and Outlook actually decide what goes to spam?

Nobody outside those companies knows the full recipe. What is public falls into four buckets:

Authentication. Is the mail provably from the domain it claims? SPF, DKIM and DMARC answer that; we cover the mechanics in SPF, DKIM and DMARC explained for sales teams. Sender reputation is the history of your domain, IPs and mailboxes. Recipient behavior means replies, deletions unread, rescues from spam and, above all, the "report spam" button. Content and structure covers words, yes, but also links, images, HTML weight and whether thousands of near-identical messages arrive at once.

The 2024 bulk-sender rules show where the weight sits. Since February 2024, Google has defined bulk senders as those sending roughly 5,000 or more messages a day to Gmail addresses, and its Email sender guidelines expect them to authenticate with SPF, DKIM and DMARC, offer one-click unsubscribe on marketing mail, and keep the user-reported spam rate below 0.3%, ideally under 0.1%. Yahoo applies similar rules in its Sender Best Practices. None of these requirements mentions vocabulary. They cover identity, an easy exit and complaints.

Which words and phrases still raise risk, and why is it the context?

Some copy does draw trouble. What unites it is that it is deceptive, high-pressure or typical of abusive senders, and the pattern gets punished.

We don't tell anyone to ban "free." "I can send a free teardown of your onboarding emails" is an honest offer to one named person. The same word in capitals beside "act now," sent to ten thousand strangers, reads as mass promotion, and recipients treat it that way.

On the law: CAN-SPAM doesn't ban vocabulary. The statute, 15 U.S.C. §7704, prohibits false or misleading header information and deceptive subject lines, and requires commercial email to identify itself as an advertisement, include the sender's valid physical postal address and offer a working opt-out that is honored within ten business days. In the EU, GDPR requires a lawful basis for processing the recipient's personal data (Article 6 of Regulation (EU) 2016/679), but it isn't the only rule: Article 13 of the ePrivacy Directive and the national laws implementing it govern unsolicited email directly, and they differ by country. Our guide to GDPR-compliant cold outreach covers both.

Why can formatting, links and tracking hurt more than any single word?

Picture a first email with branded HTML, a logo, four social icons, three links, a shortened booking URL and an open-tracking pixel. Now picture a four-sentence plain-text note from a person. Both could contain the word "pipeline." Only one resembles a newsletter blast, and a blast is what recipients are quickest to report.

Links deserve the most suspicion. Shortened URLs hide the destination, which abusive senders love. A tracking domain shared with other customers of your sending tool inherits their reputation, good or bad. A link whose visible text names one domain while the URL points to another looks like phishing.

Our defaults for a first touch: plain text or minimal HTML, zero or one link, no attachment, no tracking pixel you don't need. If you track clicks, use a custom tracking domain on your own subdomain so the reputation is yours alone.

How does sending the same template at scale turn harmless copy into a spam signal?

Fingerprinting is old technology in open-source filtering: SpamAssassin's optional Razor, Pyzor and DCC plugins checksum message bodies and ask whether the same fingerprint is being reported elsewhere. Gmail and Outlook don't say whether they do likewise, but identical content in volume being a bulk signal is hard to avoid in any serious filter.

Now the arithmetic, with a stated assumption: you send 1,000 emails a day to Gmail addresses. At the 0.3% line, three spam reports a day is trouble. At Google's recommended 0.1%, it's one. A template tweak that makes four more people per thousand click "report spam" matters far more than any word you scrubbed.

The fix is relevance. Segment tightly and make the first line say something true about that company or role, so your email isn't a duplicate of the one in the next inbox. Our piece on B2B buying signals covers how to pick that line.

Why do recipient complaints and replies outweigh word choice?

They are the only signals that measure what the filter is trying to predict: does this person want this mail? We think of a reply as a deposit into your reputation and a spam report as a withdrawal, and the withdrawal is bigger. That is why a blunt, slightly salesy email to a well-matched buyer can reach the inbox while a lovingly scrubbed one sent to a badly matched list lands in spam. The filter isn't grading your copy. It's grading your list.

So the most effective "spam word removal" happens upstream: verify addresses, target people who plausibly have the problem, make the unsubscribe path obvious so an unwanted email costs a click rather than a report, and stop following up with people who've gone quiet. Our follow-up cadence guide covers when to stop; the deliverability checklist covers the rest.

How can you test whether a phrase is really hurting your inbox placement?

Don't guess. Run a split:

1. Change one thing. Take your current email (A) and a version with only the suspect phrase changed (B). Edit the subject, opener and link at once and you learn nothing.
2. Split randomly, same list, same window. Draw both from one segment, send from the same mailboxes in the same hours, and alternate recipients rather than sending A Monday and B Tuesday.
3. Add seed addresses. Put your own test accounts at Gmail, Outlook and Yahoo into both arms and note each folder. A handful of seeds can flag a big gap, such as B landing in spam everywhere A reaches the inbox, but can't prove a small one.
4. Judge on replies, bounces and unsubscribes. Gmail doesn't hand you per-message complaint data, and placement is hard to observe at scale, so replies are the outcome you can count.
5. Read Postmaster Tools as a trend. Google Postmaster Tools reports at domain level and only with enough Gmail volume, so you can't split it by variant. Use it to confirm nothing deteriorated during the test.

How much volume? Assume two arms, 95% confidence, 80% power. To see a reply rate move from 3% to 5% you need roughly 1,500 recipients per arm. Spam-folder placement can't be read off real recipients, so that comparison needs a seed-list or placement-testing tool: to see the spam-folder rate move from 5% to 10% you need roughly 435 seed placements per arm, far more than your own few test accounts. Below those sizes, treat differences as noise. If A and B perform alike, keep the phrase you prefer.

How should you rewrite salesy cold email copy without stripping out the offer?

Keep the offer; remove the pressure and vagueness around it. Take a subject line like "FREE guaranteed 3x pipeline, act now!!!" paired with an opener like "We are the leading provider of revolutionary outbound solutions." Nothing in either can be checked, and the pair could go to any company on earth.

A rewrite: subject "Question about your Q4 outbound at Acme"; opener "I saw you're hiring two SDRs. We run outbound for teams in that spot, and I can send a free teardown of your current sequence. Worth a look?" The offer survives, the "free" survives, and the claim shrank to something deliverable. What went was the hype, the shouting, the false urgency and the filler.

Our editing rules: every claim is verifiable or something you'll show, one ask per email answerable in a line, and read it aloud. If you wouldn't say it to a stranger at a conference, don't send it to a thousand. We build an AI SDR that writes and sends every email and LinkedIn touch, so we do this edit for a living.

FAQ

Is there an official list of spam trigger words from Gmail or Microsoft?

No. Google, Microsoft and Yahoo publish none. Their filtering is largely machine-learned and driven by reputation, authentication and how recipients treat your mail, so any list online is someone's guess or a leftover from older rule-based filters.

Will a single word like "free" or "guarantee" send my cold email to spam?

Almost never by itself. They get riskier when surrounded by hype formatting, many links, mass-identical copy or an offer the recipient doesn't want. A true, specific sentence to a well-matched buyer that happens to say "free" is not what filters are built to catch.

Do spam checkers like Mail-Tester or SpamAssassin scores predict Gmail inbox placement?

Only roughly. SpamAssassin is open-source software with its own rule scores, and checkers like Mail-Tester are useful for catching broken authentication, missing headers or sloppy formatting. But Gmail and Microsoft don't filter with SpamAssassin, so a perfect score doesn't promise the inbox and a low one doesn't mean spam.

Should I avoid links and images in the first cold email?

Our default is yes: plain text, no images, at most one link, no attachment. A short reply-ready note looks like what it is, a message from a person. If you need a link, use your own domain, not a shortener.

Does personalization help cold emails avoid spam filters?

Indirectly. Filters don't reward "Hi {{first_name}}" as such. Relevance helps because relevant mail earns replies and avoids reports, and a first line that differs per recipient breaks the identical-template pattern. Token swaps on a generic body change little.

Can AI-written cold emails get flagged as spam just for being AI-written?

We have no evidence that Gmail or Outlook flag mail simply because AI wrote it, and neither has said it does. What gets flagged is the behavior cheap generation makes easy: huge volumes, near-identical bodies, poorly matched lists. A specific AI draft sent at sane volume is judged like any other email.