Ask most email marketers what causes spam filtering, and they will say something about spam words. Avoid certain phrases and your email lands in the inbox. Use them, and it goes to spam.
This model was partially true in 2005. It is not how modern spam filtering works.
Content is one factor in a multi-signal evaluation system. For established senders with good domain reputation, content signals rarely determine inbox placement. For new senders or those with marginal reputation, content signals carry more weight. Understanding exactly how content is evaluated and what actually matters versus what does not lets you focus effort on the right things.
The Role of Content in Modern Spam Filtering
Modern spam filtering at Gmail, Outlook, and Yahoo uses machine learning models trained on billions of emails. These models evaluate content as one signal among many alongside sender reputation, authentication status, recipient engagement history, and real-time sending behaviour.
For senders with High domain reputation, content signals are weak modifiers. The strong positive reputation signal from a well-established domain provides enough trust credit that moderate content risk factors are absorbed. A promotional subject line from a domain with High reputation typically lands in the inbox (or at most the Promotions tab).
For senders with Medium or Low reputation, content signals carry more weight. When the reputation signal is neutral or negative, the content score becomes a more significant tiebreaker. Content that might be ignored for a trusted sender contributes materially to filtering decisions for a marginal one.
For senders with no reputation history, new domains in warm-up content signals matter most of all. With no positive history to offset any negative content signal, the content evaluation layer has greater influence over early placement decisions.
This context determines how much attention to pay to content optimisation at any given programme stage.
What Spam Filters Actually Evaluate in Your Email
Modern spam filters evaluate content across several dimensions. Understanding each one helps prioritise where to focus.
Subject line signals. The subject line receives specific attention because it is the most visible element and has historically been abused by spammers. Patterns evaluated include: all-caps words, excessive punctuation (multiple exclamation marks), urgency language without specific context, and certain financial terminology patterns.
Body content patterns. The filter evaluates the overall language pattern of the email body. It is not looking for specific words; it is looking for patterns associated with spam templates. These include: repetitive claims about financial benefit, excessive urgency and scarcity language, certain combinations of action-demanding phrases, and content that is primarily or entirely promotional with no informational value.
HTML structure. The technical structure of your email’s HTML is evaluated alongside the content. Unusual or obfuscated HTML, extremely high image-to-text ratios, hidden text elements, and malformed header structures all add to content spam scoring.
Link analysis. Every link in your email is checked against reputation databases. Links to domains with poor reputation, links that redirect through multiple URLs, or links to unusual top-level domains receive negative scoring.
Text rendering. What the email looks like when rendered is different from what the raw HTML contains. Filters evaluate both. An email that renders as mostly images with minimal text looks like a spam avoidance technique to filters that have learned this pattern from historical spam.
The SpamAssassin Rule System
SpamAssassin is an open-source spam filter that many mail servers, corporate email systems, and smaller ISPs still use. Unlike the machine-learning systems at Gmail and Outlook, SpamAssassin operates through rule-based scoring.
Each rule adds positive or negative points to a total score. A score above the server’s configured threshold (typically 5.0, sometimes 3.0) triggers spam classification.
Environments still use it or similar rule-based systems. Second, the rules encode patterns that are consistently associated with spam and are still worth avoiding even when the filter is more sophisticated.
Common SpamAssassin rule triggers and their approximate point values:
- SUBJECT_ALL_CAPS: subject line entirely in caps (+2.0 to +3.0 points)
- MISSING_HEADERS: email missing standard headers (+2.0)
- HTML_MESSAGE_ONLY: email has HTML but no plain text alternative (+0.5 to +1.0)
- NO_RECEIVED: missing Received header in legitimate-style send (+2.0)
- MIME_HTML_MOSTLY: HTML content with very little text (+0.8)
- URI_DNSBL: a link domain is on a DNS blocklist (variable)
Run your email through mail-tester.com before any major send to see its SpamAssassin score and exactly which rules are triggering.
Content Signals That Genuinely Increase Spam Scoring
These content patterns have consistent and documented effects on spam scoring. They are worth addressing.
All-caps words in the subject line. This is consistently the highest-scoring SpamAssassin trigger for subject lines. It also affects Gmail’s subject line pattern analysis. Replace ALL CAPS with standard title case or sentence case.
Excessive exclamation marks. Multiple exclamation marks in combination with other promotional signals add to content scoring. Single exclamation marks are generally fine for warm, engaged audiences. Avoid multiples.
Missing or minimal plain text alternative. Many spam emails include only HTML, not a plain text version. Spam filters learned this pattern. Including a plain text alternative (most ESPs generate this automatically) reduces content scoring.
Deceptive subject line tactics. Fake “Re:” and “Fwd:” prefixes that imply a prior conversation that did not occur score negatively. They are also against Gmail’s guidelines. Do not use them.
Very high image-to-text ratio. If your email is primarily images with minimal readable text, spam filters register the pattern associated with image-spam avoidance techniques. Include substantive text alongside images.
Links to poor-reputation domains. Every domain your email links to is checked. This includes your tracking domain, your redirect domains, and your destination URLs. Check all domains involved in your link chain for blacklist status before sending.
Content Signals That Are Overrated as Spam Triggers
These are widely cited as spam triggers but have minimal or no direct effect on modern filtering for established senders.
The word “free.” Using the word “free” does not cause spam filtering. The context around it matters. “Free delivery on orders over $50” in an authenticated email from a well-reputed domain is not a spam signal. It is context that spam filters correctly interpret as promotional but legitimate.
Dollar signs and price references. Mentioning prices, discounts, and currency is normal commercial email behaviour. Spam filters do not flag currency symbols in isolation. They flag them when combined with other high-scoring patterns in low-reputation sending contexts.
The phrase “click here.” This is a weak link anchor text signal (good anchor text is descriptive). But it does not directly cause spam filtering for established domains.
Long emails. Email length is not a meaningful spam filter signal. Newsletters can be 5,000 words long. Transactional emails can be 50 words. Neither length is inherently more or less likely to be filtered.
Bright colours and multiple fonts. HTML formatting choices are a weak content signal at most. The high image-to-text ratio concern is real. Colourful formatting itself is not.
HTML and Technical Content Signals
Technical HTML quality affects both spam scoring and email rendering.
Plain text alternative. As noted above, including a plain text version reduces spam score and improves rendering for email clients that prefer or require plain text.
Clean HTML structure. Malformed HTML, unclosed tags, invalid nesting, and duplicate style definitions are associated with automated spam generation tools. Clean, valid HTML reduces this signal. Most ESPs generate valid HTML through their template builders, but custom HTML should be validated before sending.
Balanced image-to-text ratio. The guideline is at minimum 30% readable text content relative to the total email. For image-heavy designs, add text content in image alt tags and supplementary text sections.
Tracking domain reputation. Most ESPs rewrite links in emails to go through a tracking domain. If your ESP uses a shared tracking domain and that domain has reputation problems from other senders, your emails inherit those problems. For high-volume senders, configuring a custom tracking domain ensures link tracking reputation is your own.
Email header completeness. Legitimate email has a complete set of standard headers: Date, From, To, Subject, Message-ID, MIME-Version. Missing standard headers trigger spam scoring in rule-based filters. ESP-generated email typically includes all required headers automatically.
Link and Domain Reputation in Email Content
Every URL in your email connects to a domain. That domain has reputation signals that are evaluated as part of content scoring.
What Is Checked
Spam filters check the domains your links point to against DNS-based reputation lists (DNSBLs). The most widely integrated are Spamhaus DBL (Domain Block List) and SURBL. A link to a domain on either of these lists adds significant negative content scoring.
This check applies to all domains in your link chain not just the final destination. If you use a redirect service that goes through an intermediate domain before reaching your website, that intermediate domain is also checked.
Your Tracking Domain
If you use click tracking in your ESP, every link in your email is rewritten to go through a tracking URL. The domain in that tracking URL (your-esp.com/track/… or click.yourdomain.com/…) is evaluated by spam filters.
Shared tracking domains on large ESPs sometimes accumulate reputation issues from other senders on the same tracking domain. For high-volume senders, setting up a custom tracking domain (click.yourdomain.com) isolates your link tracking reputation from the shared pool.
Checking Link Domains Before Sending
Before major campaigns, check your primary sending domain, your tracking domain, and any external domains you link to with MX Toolbox at mxtoolbox.com/blacklists.aspx.
Content Strategy for Different Reputation Levels
The appropriate level of content concern depends on your current domain reputation.
High reputation senders. Content signals have minimal impact on your inbox placement. Your primary focus should be content quality and engagement what drives opens, clicks, and replies. Avoid the most obvious triggers (all-caps, missing plain text) as a matter of best practice, but do not overthink content-level spam avoidance.
Medium reputation senders. Content signals carry more weight when reputation is neutral. During reputation recovery, avoid any content pattern that adds meaningful SpamAssassin points. Run every campaign through mail-tester.com before sending. Aim for a score below 1.5.
New domain / warm-up senders. Content signals matter most here. Use conversational, plain-text-style content during warm-up rather than heavily formatted HTML campaigns. Personal, direct communication generates better engagement signals. Strong engagement during warm-up builds positive reputation faster than promotional content that generates deletion-without-open signals.
Low or bad-reputation senders. Content is a secondary concern. The primary issue is reputation. Focus entirely on reputation recovery first: verified lists, high-engagement segments, reduced volume, before worrying about content optimisation.
Key Takeaways
- Content is one factor in multi-signal spam filtering. For senders with High reputation, content signals rarely determine inbox placement. For new or marginal-reputation senders, content matters more.
- SpamAssassin is still used by many corporate and smaller ISP email systems. Its rule-based scoring system penalises specific patterns: all-caps subject lines, missing plain text alternatives, deceptive Re:/Fwd: prefixes, and high image-to-text ratios.
- Genuinely impactful content signals: all-caps subject words, missing plain text alternatives, fake reply prefixes, very high image-to-text ratio, and links to domains on spam blocklists.
- Overrated content triggers: individual words like “free,” currency symbols, the phrase “click here,” email length, and formatting choices like colours and fonts.
- Every domain your email links to, including tracking domains and redirect intermediaries, is checked against reputation lists as part of content evaluation.
- Content strategy should match your reputation level. High-reputation senders focus on engagement quality. Medium or recovering senders minimise content risk. New senders prioritise plain, conversational content that generates strong engagement signals.
Frequently Asked Questions
No. Modern spam filters consider the overall email, sender reputation, engagement, and sending patterns rather than blocking individual words.
Not by itself. Plain text can encourage replies during warm-up, but it does not guarantee better deliverability.
Test new templates and major content changes. For established campaigns, monthly testing is usually enough.
Yes, but their impact is usually smaller than sender reputation, engagement, authentication, and list quality.
Maintain a strong sender reputation, authenticate your domain, keep your list clean, and send relevant content consistently.
Conclusion
Content matters for email deliverability but not in the way most advice suggests. It is not about avoiding specific words. It is about avoiding patterns that add meaningful spam scoring, maintaining technical HTML quality, and ensuring every domain in your email’s link chain is free from blacklist issues.
For established senders, content quality drives engagement signals, which is where the real deliverability impact of content decisions lies. For new or marginal-reputation senders, content risk minimisation is worth genuine attention.
