Spam Triggers in Cold Email: Words, Formatting and Patterns That Get You Filtered

The classic “527 spam words to avoid” lists are mostly folklore — modern filters weigh sender reputation and engagement far more than any single word. But copy still matters: certain phrases, formatting habits and structural patterns are statistically loud markers of bulk mail, and a cold domain without reputation to spend can't afford to wear them. This guide covers what actually trips filters in B2B outreach today, and how to say the same thing in language that reads — to machines and humans alike — as a business letter, not a broadcast.

How spam filters actually read your copy in 2026

The mental model of a forbidden-words list that instantly condemns your email is fifteen years out of date. Modern filtering at major providers is probabilistic and holistic: your domain and IP reputation, authentication results, volume patterns, recipient engagement history and content signals all feed a score, and content is one input among many. A trusted sender can write “free trial” all day; an unknown domain sending its first thousand messages gets far less benefit of the doubt.

That's precisely why copy matters most for cold email. A prospecting domain is by definition short on reputation and engagement history, so the filter leans harder on what it can see: the text, the structure, the links, and how much the message resembles known bulk patterns. Content is the tiebreaker, and cold senders live permanently in tiebreaker territory.

One framing note before the lists: everything here is about making a legitimate, individually addressed business letter look like what it is. If you're sending ten thousand identical messages to a scraped list, no vocabulary substitution will help — the sending pattern itself is the spam signal, and it should be.

Phrase categories that raise scores (and what to write instead)

Think in categories rather than memorizing word lists, because filters generalize and so should you. The reliably risky families in B2B contexts: money and prize framing (“100% free”, “guaranteed income”, “no cost to you”), pressure and urgency (“act now”, “final notice”, “limited time only”), evasiveness (“this is not spam”, “no strings attached”, “no obligation”), pharmaceutical and adult vocabulary (rarely relevant in B2B, instantly fatal when present), and inflated superlatives (“revolutionary breakthrough solution”, “world's best”).

The pattern behind the pattern: spam language promises much, demands speed and pre-emptively denies bad faith. Business language does the opposite — it's specific, verifiable and calm. “Guaranteed 300% ROI” is a spam phrase; “clients in your segment typically cut processing time by a third — happy to show the math” carries the same commercial intent with none of the statistical stink, because it's falsifiable, hedged and concrete.

Two B2B-specific traps deserve mention. First, fake familiarity — subject lines like “Re: our conversation” or “Following up on your request” when neither existed. Some filters treat deceptive thread markers as a manipulation signal, recipients treat them as a lie, and under CAN-SPAM a materially misleading subject line is a legal problem, not just a stylistic one. Second, the word “free” panic: in moderation and in honest context (“free 20-minute assessment”) it's survivable for a sender with decent reputation; stacked with urgency and exclamation marks (“FREE!!! Act today!”) it's a caricature of the exact pattern filters were built to catch.

• Money promises: “100% free”, “guaranteed returns”, “pure profit” → name a concrete, hedged outcome with real numbers
• Urgency theater: “act now”, “expires today”, “last chance” → give a real reason for timing, or drop the deadline
• Self-denial: “this is not spam”, “no obligation whatsoever” → delete; legitimacy is shown, not declared
• Superlatives: “revolutionary”, “world-leading”, “game-changing” → one specific capability plus one verifiable proof point
• Fake threads: “Re:”, “Fwd:” on a first touch → honest subject naming the topic, ideally with a company-specific detail
• Pressure CTAs: “click here now” → a plain question that invites a reply, not a click

Formatting red flags that outweigh any word choice

Filters read typography as behavior. ALL-CAPS runs, multiple exclamation marks, rainbow-colored or oversized fonts, and excessive bolding are the digital equivalent of shouting through a megaphone — and they score accordingly. The same goes for currency symbols stacked with numbers (“$$$ SAVE $5,000 $$$”) and emoji-dense subject lines.

Structure signals matter as much as decoration. A healthy business email is mostly text with at most a couple of links; a message that's one big image (a common trick to hide text from filters — and filters know it), or that carries five tracked links, a button, a banner and a signature graphic, matches the commercial-broadcast template. Link shorteners are a specific hazard: bit.ly-style domains are heavily abused by spammers, so their reputation taxes yours. Link the real destination, on your own domain, with the URL visible for what it is.

Attachments on a first touch deserve their own warning. An unsolicited PDF — let alone a .zip or anything executable — triggers both machine filters and human security training; corporate recipients are literally drilled to distrust exactly that message. If you have a document worth sharing, offer it in the reply (“happy to send over the two-page breakdown”), which is simultaneously safer for deliverability and a better conversion device, since it gives the prospect an easy yes.

Structural patterns: what makes a message smell mass-produced

Beyond words and formatting, filters — and increasingly, recipients — detect industrial production. The strongest structural signal is textual similarity at scale: hundreds of messages that are identical except for a swapped first name, sent in a burst from one domain, form a fingerprint no vocabulary can disguise. Genuine per-recipient variation — different first lines, different order of points, sentences that only make sense for that one company — is both the best personalization practice and the best structural defense, because personalized mail is statistically unlike bulk mail.

Length and shape play a role too. Cold emails converging on the same 40-word growth-hack template ('name-drop, one-liner pitch, “worth a chat?”') are a genre that filters and humans now pattern-match instantly. On the other end, a 600-word essay with twelve links resembles a newsletter. The healthy zone for a first B2B touch sits around 60–150 words: enough to be specific about them, short enough to respect a stranger's attention, and structurally unlike both spam genres.

Finally, mind the mismatch signals: a subject line about “your logistics costs” over a body pitching HR software; a sender name that doesn't match the signature; a plain-text opening followed by a heavily branded HTML footer. Inconsistency reads as assembly-line output where different modules were bolted together — because that's usually what it is. One voice, one topic, one coherent visual register per message.

How to test your copy before it costs you reputation

Don't guess — test, in ascending order of realism. Start with a plain read-aloud: if a sentence would sound absurd said to the prospect's face across a table, it's spam-adjacent by definition. Then run the mechanical checks: spam-score tools based on classic rulesets catch the crude stuff (caps, exclamation density, notorious phrases) cheaply, and while they don't reflect modern reputation-weighted filtering, a message that fails them will certainly fail the real thing.

The meaningful test is seed inboxing: send the actual message, from the actual sending mailbox and infrastructure, to a panel of test addresses across the providers your list actually uses — for B2B that means both consumer-grade mailboxes and accounts behind common corporate gateways. Check where it lands, not just whether it arrives. Do this for every new template and after any significant copy change, because content scoring interacts with your current reputation and yesterday's clean result doesn't cover today's edit.

Isolate variables when a test fails. Send the suspect copy from a warmed, healthy mailbox: if it inboxes there but spams from your campaign domain, your problem is reputation, not words. Strip links and re-test: if placement jumps, one of your destinations or trackers is the poison. This kind of bisection takes twenty minutes and beats a week of superstitious word-swapping.

The mindset fix: write letters, not broadcasts

Step back from the lists and the pattern is almost embarrassingly simple: everything that trips filters is a symptom of writing to a crowd, and everything that passes is a symptom of writing to a person. Urgency theater exists because broadcasts can't wait for a reply; superlatives exist because generic pitches have nothing specific to say; image-heavy templates exist because newsletters need branding. Remove the crowd, and the triggers remove themselves.

That's the operating logic of address-based B2B outreach: a finite list of companies where your offer is genuinely relevant, a researched reason to write to each decision-maker, 60–150 words of plain text, one honest question, a real signature with a real address, and a working opt-out. Such a message has nothing for a content filter to seize on — and just as important, nothing for a wary human to seize on either. In B2B, the recipient's own “report spam” click, driven by irrelevance rather than vocabulary, does more reputational damage than any word on any list.

So use this guide as a linter, not a strategy. Run your drafts against the categories, fix the formatting flags, test on seeds — and then invest the remaining time where the leverage actually lives: tighter targeting, better research, and copy specific enough that no filter, silicon or human, could mistake it for bulk.

• Would I say this sentence to the prospect's face? If not, cut it
• Zero ALL-CAPS words, at most one exclamation mark per email — ideally none
• One or two links maximum, full URLs on your own domain, no shorteners
• No attachments on first touch; offer documents via reply instead
• Subject line honest, specific, no fake Re:/Fwd:, under ~50 characters
• 60–150 words of mostly plain text; recipient-specific first line
• Real sender identity, physical address and working opt-out in the footer
• Seed-tested from the production mailbox before the sequence launches

FAQ