Email copy that hurts deliverability: spam words, subject lines, links and images

When mail lands in spam, the first instinct is to rewrite the copy. It's usually the wrong place to start. In the cases we investigate, authentication, sending reputation and list quality cause far more filtering than wording does, and why transactional email goes to spam covers the order to check them in.

Copy still matters in two ways. Filters look for deception and for the structures spammers rely on. And readers who feel tricked press Report spam, which raises the spam rate Gmail asks you to keep below 0.3%.

The spam-word myth

Lists of hundreds of "spam trigger words" circulate widely. Gmail's sender guidelines, Yahoo's sender best practices and Microsoft's requirements for high-volume senders don't contain one. What they spell out is authentication, complaint rates, unsubscribe handling and accurate headers.

Word rules do exist in some filters. SpamAssassin, the open-source filter some companies and hosts run on their own mail servers, has phrase rules aimed at scam wording: "Dear Friend", "one hundred percent guaranteed", unclaimed money. Each adds at most a few points toward a default spam threshold of 5, and none of its phrase rules reaches that threshold alone. A product email that mentions free shipping isn't your problem.

Two things about words do matter:

• Misspelling words to slip past a filter (fr3e, V1agra) backfires. Filters look for obfuscation, and SpamAssassin ships rules for exactly that.
• Words shape how readers react. A subject that oversells gets the open, disappoints, and gets reported.

Subjects and From names that mislead

This is where the mailbox providers are explicit. Gmail's guidelines say the subject, headers and display name should accurately represent the sender and the content, and they give examples:

• Don't start a subject with Re: or Fwd: unless the message is a real reply or forward.
• Use the display name to identify the sender. Gmail lists names like "URGENT REQUEST" and "Last Chance" as ones to avoid, because they put subject content where the sender belongs.
• Don't put the recipient's name in the display name, or imply a thread, as in "User (2)".
• Don't put emojis or images next to a display name or brand name to imply it has been verified.

Yahoo tells senders not to use false or misleading header information or deceptive subject lines. Microsoft asks for accurate subject lines, and for a From or Reply-To address that is valid, on your real sending domain, and able to receive replies. In the US, the FTC's CAN-SPAM compliance guide forbids misleading header information and deceptive subject lines in commercial email.

A marketing email dressed as a receipt or an invoice ("Your order #4821") breaks the same rule. The reverse is a problem too: Gmail says not to mix content types, such as promotions inside a sales receipt.

ALL CAPS, exclamation marks and emoji

Shouting is a weak filter signal and a strong reader signal. In SpamAssassin's published scores, a subject written entirely in capitals adds 0.5 points, and a subject with both ! and ? adds 0.1. Those numbers won't sink a message. The cost is with readers: a subject like "LAST CHANCE!!!" looks like the spam they already delete, and some will report it.

Emoji are the same kind of decision. Gmail's guidelines mention them in two places: don't use them to imitate graphic elements, such as a check mark that suggests a verified brand, and keep them out of the display name. Neither is a rule against an emoji in a subject line, so treat it as a question of style and trust. One that fits your brand is fine. A row of them in every subject reads as a sender competing for attention, and they can render as empty boxes where a font lacks them.

Links: shorteners and mismatched domains

Gmail asks that links in the body be visible and easy to understand, so recipients know what to expect when they click. Public URL shorteners work against that, and they make you share a reputation with everyone else who uses the same service. Spamhaus says in its Domain Blocklist FAQ that spammers frequently use URL shortening services to try to avoid spam filtering. SpamAssassin ships shortener rules too, including one for links a shortener has disabled for abuse.

A shortener rarely decides delivery on its own. The DBL gives abused shorteners and redirectors their own return code, and Spamhaus set it up that way so filter operators could choose whether to block those links or only score them. But in an email a shortener buys you nothing, since the reader sees your link text, not the URL. Use full links on your own domain.

Two more habits, both our advice:

• If the visible link text is a URL, link to that URL. Text that shows yourbrand.com while the link goes somewhere else is how phishing looks, and readers who hover will notice.
• Keep link domains close to your From domain. A message from yourbrand.com that links to five unrelated domains is harder for a filter or a person to trust.

Image-only emails and hidden text

An email that is one large image, or a few images with a line of text, gives a filter little text to read. SpamAssassin has a family of HTML_IMAGE_ONLY rules for HTML that is mostly images with little text, scored up to about 2.8 points. Many readers won't see the images either: Outlook blocks automatic picture downloads by default, so an image-only email opens as empty frames until the reader chooses to download pictures.

Put the message in live text and let images support it. Give every image alt text that says what it shows, and read the email with images turned off before you send.

Don't fix a thin message by hiding text in the HTML for filters to find. Gmail's guidelines say not to use HTML and CSS to hide content, and warn that hiding content might cause messages to be marked as spam.

Send a plain-text part

A well-formed HTML email is a multipart/alternative message: an HTML part plus a plain-text version of the same content (RFC 2046, section 5.1.4). Gmail's guidelines don't require the text part, and leaving it out is a small signal on its own: SpamAssassin's MIME_HTML_ONLY rule scores 0.1. A text part that doesn't match the HTML scores more (MPART_ALT_DIFF), so generate it from the same content instead of pasting a placeholder. Text-only mail clients show it, and it costs nothing to send.

Before you send

• From name: your brand, or a real person at it, on an address where replies reach someone.
• Subject: says what's inside. Re: and Fwd: only on real replies and forwards.
• Links: on your own domain, no public shorteners, link text that matches the destination.
• Body: readable with images off, with a plain-text part generated from the same content.
• Content type: promotions stay out of receipts and account mail.