Content Choices That Decide Whether Your Cold Email Clears Spam Filters

Two emails leave the same warmed-up mailbox to the same company: one lands in the inbox, the other in spam. Infrastructure identical — the difference was in the message body. Content is the deliverability lever writers control directly, and in cold B2B outreach the rule is simple to state and hard to fake: the more your email looks and behaves like genuine one-to-one business correspondence, the better it gets treated. Here is what that means at the level of links, images, formatting and words.

What filters actually read in your message

Modern spam filtering is not a checklist of forbidden words. Providers score messages with statistical models trained on billions of examples of what users mark as spam versus what they read and answer. The model sees everything: the subject line, the text, the HTML structure, the number and destination of links, the text-to-image ratio, attachments, even how similar this message is to thousands of others arriving from the same sender.

That training data creates two recognizable profiles. Marketing blast: heavy HTML, banner images, multiple links, tracking pixels, promotional phrasing, hundreds of near-identical copies. Business correspondence: plain or lightly formatted text, a few sentences, zero or one link, a signature, each message unique. Your cold email will be scored by which profile it resembles — and everything in this guide is a way of moving your message toward the second profile honestly.

The word honestly matters. This is not about tricking filters into delivering bulk mail — tricks decay within months as models retrain, and gaming filters for mass sends is the definition of spamming. It is about making sure a legitimate, targeted note to a specific decision-maker is not dressed up in the costume of a promotion. If your outreach is genuinely one-to-one in intent, your job is just to stop the packaging from contradicting that.

Links: the heaviest content signal

Links are weighted heavily because they are how spam monetizes. Each additional link in a message raises its score, and certain link properties raise it sharply: URL shorteners like bit.ly (reputation destroyed by years of abuse), raw IP-address links, mismatches between anchor text and destination, and links to domains with poor reputation.

For a cold first touch, the strong default is zero or one link. Zero is perfectly viable — a first email whose only call to action is a reply question needs no link at all, and reply-based CTAs both convert well in B2B and generate the reply signal that builds sender reputation. If you include one link, make it to your own domain, with honest anchor text, placed once.

If your platform rewrites links for click tracking, configure a custom tracking domain — a subdomain of your sending domain — so the rewritten URL still belongs to you. Links routed through a shared platform tracking domain inherit the reputation of every other customer using it, which is a lottery you do not need to play. And think twice about whether click tracking earns its cost on a two-sentence email whose goal is a reply: many senders drop it for cold sequences and lose nothing they were actually using.

Images, attachments and HTML weight

Images in a cold email are almost pure downside. They add spam weight (image-heavy, text-light messages are a classic spam pattern), they often fail to render because corporate mail clients block remote images from unknown senders by default, and they signal marketing blast rather than a colleague's note. Logos in signatures, banners, product screenshots — save them for later in the conversation, after a reply has established the relationship.

Attachments on a first touch are worse. Unsolicited attachments are the primary malware delivery mechanism, and both gateways and human recipients treat them accordingly. A PDF one-pager attached to a cold email lowers delivery odds and looks presumptuous. If you have material worth sharing, mention that it exists and offer to send it — that offer is itself a natural reply-generating CTA.

HTML weight is the subtler issue. Emails composed in template builders carry hundreds of lines of nested table markup, inline CSS and tracking scaffolding — a structure filters instantly recognize as bulk-mail tooling, because no human writes email that way. Send plain text or minimal HTML: paragraphs, maybe bold on one phrase, a simple text signature. As a rule of thumb, if the HTML source of your email is several times longer than its visible text, you are sending a flyer, not a letter.

Wording: patterns, not magic words

Lists of forbidden spam trigger words circulate endlessly, and they mislead by oversimplifying. No modern filter blocks an email for containing the word free. What raises scores is the promotional pattern those lists gesture at: urgency and pressure (act now, limited time, last chance), hype (guaranteed, incredible, revolutionary), money-out-of-nowhere framing (risk-free, no obligation, save 50%), and typographic shouting — ALL CAPS, multiple exclamation marks, currency symbols scattered through the text.

One such phrase in an otherwise normal business message costs little. Several stacked together shift the whole message into the promotional profile — and they do double damage, because the same phrases also lower reply rates with human readers. B2B decision-makers have spent years developing the same classifier the filters use: anything that sounds like an ad gets deleted unread.

The reliable test is to read your draft aloud and ask whether you could send it, word for word, to a professional acquaintance without embarrassment. 'Don't miss this exclusive opportunity to revolutionize your workflow!!' fails that test. 'Saw you're hiring three warehouse managers — usually a sign the current process is straining. We might be able to help with part of that.' passes it. Write the second kind of sentence and the trigger-word question mostly answers itself.

Subject lines follow the same logic compressed into one line: lowercase or sentence case, specific, no brackets-and-emoji decoration, no RE: or FW: faked onto a first touch. Deceptive subject lines are not just a filter risk — under CAN-SPAM, misleading subject lines and false header information are explicitly prohibited, and under GDPR-era practice, transparency about who you are and why you are writing is part of the legitimate-interest basis most B2B outreach relies on.

Structure of a first touch that passes both filters and humans

Everything above converges on a compact format that consistently performs in address-based B2B outreach. It is short — 60 to 120 words. It opens with the reason this specific company and person, which is both persuasion and a uniqueness signal. It makes one concrete relevance claim with a number-as-range rather than a hyped promise. It ends with a low-friction reply question. It closes with a real signature: name, role, company — the identification a legitimate business sender is expected to provide.

Personalization is doing quiet deliverability work throughout. When the first line references the recipient's actual situation, every message in the campaign differs from every other, and the near-duplicate detection that catches template blasts has nothing to latch onto. This is the structural advantage of targeted outreach over volume spam: the thing that makes the message persuasive is the same thing that makes it deliverable.

One more structural element: the opt-out. A blunt unsubscribe link styled like a newsletter footer is one option, but in low-volume B2B outreach a plain sentence — 'If this isn't relevant, tell me and I won't write again' — does the same job, reads human, and generates a reply signal even from people declining. Whichever form you choose, honoring it immediately and permanently is a legal requirement under CAN-SPAM and the practical core of GDPR compliance, and every ignored opt-out is a future spam complaint.

Common content mistakes and a pre-send check

The recurring failures are worth naming because they are all self-inflicted and all fixable in minutes. Sending the flyer: a designed HTML template with logo header and three CTA buttons, to someone who never subscribed. Link stuffing: website, case study, calendar, LinkedIn, all in one message. The fake thread: RE: in the subject of a first touch — a deception that filters increasingly detect and recipients always resent. The essay: 400 words explaining the whole product, guaranteeing the message is skimmed and deleted. And the invisible blast: two hundred identical messages with only the first name swapped, which is template spam wearing a name tag.

Before any campaign goes out, run the draft through a short content check. It takes two minutes and catches the majority of content-level deliverability problems before they cost you reputation.

• Length 60–120 words, readable in under thirty seconds.
• Zero or one link; if tracked, on your custom tracking domain; no shorteners.
• No images, no attachments in the first touch.
• No urgency/hype stack; no ALL CAPS; at most one exclamation mark, preferably none.
• Subject line specific, plain, no fake RE:/FW:, no emoji decoration.
• First line is unique to the recipient, not a swapped-in variable on a generic opener.
• Real signature with name, role and company; working opt-out path stated or linked.
• Test send to your own mailboxes across Gmail and Outlook before launch.

FAQ