Content Psychology & Filtering Signals

The Content Myth

One of the most persistent myths in email deliverability is that specific words or phrases cause spam placement. “Use of the word ‘free’ triggers spam filters.” “All caps in subject lines send you to junk.” “Too many images flag your email.”

These beliefs are rooted in how spam filtering worked 20 years ago. Legacy filters like SpamAssassin used pattern matching and keyword scoring — messages received points for containing “free,” “buy now,” “act fast,” and other phrases commonly found in spam.

Modern filtering has moved far beyond keyword matching. Gmail, Microsoft, Yahoo, and other major providers use machine learning models trained on billions of messages. These models evaluate the overall structure, behavior patterns, and reputation signals associated with a message — not individual words.

Content matters. But it matters far less than most senders think, and in different ways than most expect.

How Content Actually Affects Filtering

What Matters: Structural Patterns

Modern content analysis looks at the overall structure of a message rather than specific words:

HTML quality and complexity:

• Excessively complex HTML (deeply nested tables, inline CSS overuse, broken tags) can be a minor negative signal
• HTML that closely resembles templates used in known spam campaigns is flagged by pattern matching
• Clean, well-structured HTML is neutral — it doesn’t help, but at least it doesn’t hurt

Text-to-image ratio:

• Image-only emails (no text at all) are suspicious because spammers use images to avoid text-based analysis
• A reasonable balance of text and images is fine
• The “golden ratio” myth (exact percentages of text vs. images) is not supported by evidence

URL reputation:

• Every URL in your message is checked against reputation databases (Google Safe Browsing, URIBL, SURBL)
• A single link to a domain on a URL blocklist can tank an otherwise clean message
• URL shorteners (bit.ly, tinyurl) are frequently flagged because they obscure the destination and are heavily used in spam
• Multiple tracking redirects in a single link chain add URL inspection overhead and can appear suspicious

Attachment analysis:

• Executable files (.exe, .bat, .scr) are almost universally blocked
• Password-protected archives are flagged as potential malware delivery
• PDF and document attachments are scanned for malicious content
• Large attachments are not a spam signal per se, but they affect deliverability indirectly (slower delivery, more resource-intensive processing)

What Matters: Content-Reputation Correlation

The most important content signal isn’t in the content itself — it’s in how the content correlates with the sender’s reputation:

New content patterns from a trusted sender: When a sender with established reputation Send a message that looks different from their usual content (different template, different link patterns, unusual HTML structure), this can trigger additional scrutiny. The model expects consistency.

Content that matches spam corpus patterns: If your message structure closely resembles recently flagged spam (similar HTML, similar link patterns, similar header structure), the model assigns higher spam probability. This is content fingerprinting, not keyword matching.

Engagement prediction based on content: Gmail’s model can predict engagement based on past content performance. If messages from your domain with similar content patterns have historically low engagement, the model expects low engagement for the new message too.

What Barely Matters: Specific Words

Individual words and phrases have near-zero impact on modern filtering:

• Using “free” in a subject line does not trigger spam filters
• “Act now,” “limited time,” or “exclusive offer” are not spam triggers
• Dollar amounts, percentages, and promotional language are neutral
• ALL CAPS subjects are mildly negative (for engagement, not filtering) because they feel aggressive
• Emoji in subject lines are neutral for filtering but can affect engagement positively or negatively depending on the audience

The reason these myths persist is that they were true 15 years ago. Marketers who tested these variables saw correlations — but the correlations were with outdated filtering systems that have long since been replaced.

The Psychology of Email Content

While content doesn’t directly control spam placement, it profoundly affects the engagement signals that do. Content psychology operates at a different level than filtering — it shapes whether recipients open, read, click, or complain.

Subject Line Psychology

The subject line is the most important piece of copy in any email because it determines the open. Key principles:

Relevance over cleverness: A subject line that clearly states what the recipient will get outperforms a clever or mysterious one. “Your March account summary” outperforms “You won’t believe what happened this month.”

Personalization that adds value: Using the recipient’s name is weak personalization. Referencing their specific behavior (“Your saved item is back in stock”) is strong personalization that signals relevance.

Appropriate urgency: Genuine urgency (“Your subscription expires tomorrow”) improves opens. Manufactured urgency (“LAST CHANCE — deal ends tonight” when it doesn’t) erodes trust and increases complaints over time.

Length: Subject lines between 40–60 characters consistently perform well across devices. Mobile devices truncate longer subjects, hiding potentially important information.

Preview text (preheader): The text that appears after the subject line in the inbox preview is often overlooked. It provides a second chance to communicate value and improve open rates. Many senders leave it as default HTML boilerplate — a wasted opportunity.

From Name Recognition

The From name affects opens more than the subject line in many cases:

• Recipients decide to open or ignore based on who sent the message before they read the subject
• Consistent, recognizable From names build familiarity
• Changing From names frequently confuses recipients and can increase complaints
• Including the brand name is essential for recognition in crowded inboxes

Content Relevance and Engagement

The body content determines post-open behavior: do they click, reply, or close?

Clear value proposition: Every email should answer “why should I care about this?” within the first 2–3 sentences. Readers scan; they don’t read sequentially.

Single clear call to action: Emails with one clear CTA outperform emails with multiple competing CTAs. More choices lead to less action (the paradox of choice).

Scannable structure: Short paragraphs, bullet points, bolded key phrases, and clear visual hierarchy help readers find value quickly.

Mobile optimization: Over 60% of email opens occur on mobile devices. Content that requires horizontal scrolling, has tiny text, or relies on hover interactions fails for the majority of recipients.

Content Signals That Actually Hurt

While individual words don’t matter, several content practices genuinely increase spam risk:

Deceptive Content

Misleading subject lines: A subject that says “RE: Your account” when there was no previous conversation is deliberately deceptive. Providers increasingly detect this pattern and penalize it.

False urgency or scarcity: “Only 3 left!” when there are thousands in stock. Recipients learn to distrust these signals and engagement declines over time.

Phishing indicators: Messages that mimic login pages, ask for credentials, or imitate system alerts from services the sender doesn’t represent. These trigger specialized phishing detection models.

Technical Content Issues

Broken tracking links: When tracking redirect chains break (returning 404s or timing out), providers that follow links during scanning will flag the message.

Invisible text: White text on a white background, zero-font-size text, or text hidden with CSS. These are classic spam obfuscation techniques that modern filters detect easily.

Excessive tracking pixels: Multiple tracking pixels or beacons from different services add clutter and slow rendering. One pixel per message is standard.

JavaScript in email: Most email clients strip JavaScript entirely. Including it is pointless and triggers security filters.

Reputation-Harming Content Patterns

Affiliate links: Links to known affiliate networks used in spam have poor URL reputation. Legitimate affiliate marketing should use branded redirect domains.

Link-heavy messages: An email with 50+ links is structurally unusual and can trigger content analysis. Normal marketing emails typically have 5–15 links.

Attachment-heavy messages: Sending large attachments to large lists is unusual behavior. If you need to share documents, link to hosted files instead.

Testing Content Impact

Since content effects are subtle and context-dependent, testing is the only way to know what works for your audience:

A/B Testing Framework

Subject line testing:

• Test one variable at a time (length, personalization, urgency, question vs. statement)
• Use a statistically significant sample (at least 1,000 per variant)
• Measure open rate as the primary metric
• Run tests for a consistent time window before selecting a winner

Content testing:

• Test layout changes (single column vs. multi-column, image heavy vs. text heavy)
• Test CTA placement and copy
• Measure click rate and click-to-open rate
• Consider post-click behavior (conversions, not just clicks)

Frequency testing:

• Test different sending frequencies against randomly split cohorts
• Measure long-term engagement, not just immediate response
• Track unsubscribe and complaint rates per frequency group
• Allow enough time for fatigue effects to manifest (4+ weeks minimum)

Inbox Placement Testing

After content changes, test inbox placement:

• Send to seed accounts across all major providers
• Check that content changes haven’t triggered new filtering
• Compare placement before and after the content change
• Pay special attention to providers with aggressive content filtering (Microsoft)

The Content Hierarchy of Impact

A realistic hierarchy of what matters most for content-related deliverability:

1. Sender reputation (overwhelms all content signals)
2. URL reputation (links in your message)
3. Structural patterns (similarity to known spam templates)
4. HTML quality (broken, overly complex, or suspicious code)
5. Engagement response (how recipients react to the content)
6. Subject line accuracy (does it match the body?)
7. Visual content signals (image-only, hidden text, deceptive formatting)
8. Specific words or phrases (nearly irrelevant in modern filtering)

The practical takeaway: write content that your audience finds valuable, use clean HTML, don’t include suspicious links, and don’t try to trick or mislead. The content-based “spam triggers” that marketers obsess over are a distraction from the signals that actually matter: reputation, engagement, and list quality.