How do spam filters decide if an email is spam?

Spam filters evaluate emails across four main categories: (1) Sender reputation — the trust score of your sending domain and IP, built on complaint rates, bounce rates, and engagement history. (2) Authentication — whether SPF, DKIM, and DMARC pass. (3) Content signals — text analysis for spam patterns, HTML structure, link reputation. (4) Engagement signals — whether recipients at that ISP typically open, click, or delete emails from your domain.

Does Gmail use AI to filter spam?

Yes. Gmail uses machine learning models trained on billions of spam examples to classify email. These models analyze sender reputation, content, authentication, and user behavior signals simultaneously. Gmail claims its filters block over 99.9% of spam, phishing, and malware before it reaches users. Gmail also uses engagement data — emails from senders that users frequently engage with are less likely to be filtered.

Why does the same email go to spam at Gmail but not Outlook?

Gmail and Outlook use different spam filtering algorithms, reputation databases, and signal weights. Gmail weighs user engagement (opens, replies) and DMARC compliance heavily. Outlook's SmartScreen focuses more heavily on IP reputation and content patterns. Your domain or IP reputation may differ between the two systems, causing different filtering outcomes for identical emails.

Does email content still affect spam filtering?

Content matters, but less than it did in the early 2000s. Modern spam filters primarily use sender reputation and engagement signals — not just keyword matching. However, certain content patterns still raise risk: excessive punctuation and capitalization, high image-to-text ratio, suspicious links, missing unsubscribe, and missing physical address. Clean HTML and clear copy reduce content-based filtering risk.

What is a spam trap and how does it affect deliverability?

A spam trap is an email address used by ISPs and blocklist operators to identify senders who are not following best practices. Pristine spam traps are addresses that were never valid — hitting them means you're sending without permission. Recycled spam traps are old, abandoned addresses reactivated as traps — hitting them means poor list hygiene. Both trigger blocklisting and reputation damage.

How Email Spam Filters Work: What ISPs Check Before Delivering Your Email

Spam filters at major ISPs like Gmail, Outlook, and Yahoo evaluate every incoming email across dozens of signals before deciding whether it reaches the inbox. Understanding how these filters work is essential for diagnosing deliverability problems — and for building an email program that consistently lands in the inbox.

Email Spam Filtering Scale

99.9% of spam Gmail claims to block before reaching users

15B+ emails Gmail processes daily

45% of all email sent globally is spam Statista

0.30% spam complaint rate that triggers Gmail filtering

The Four Categories Spam Filters Evaluate

1. Sender Reputation

Sender reputation is the most important factor in modern spam filtering. ISPs maintain reputation scores for:

Sending domains: The domain in your From address
Sending IPs: The IP addresses your email actually sends from
Root domains: The organizational domain underlying subdomains

Reputation is built on:

Spam complaint rate: The percentage of recipients who mark your email as spam (the most heavily weighted signal)
Bounce rate: Invalid addresses signal poor list hygiene
Engagement rate: Opens, clicks, and replies tell ISPs that subscribers want your mail
Sending history: Consistency of volume and frequency over time

2. Authentication

Modern spam filters require authentication before they’ll even fully evaluate the content:

SPF: Does the sending IP appear in the domain’s authorized sender list?
DKIM: Is there a valid cryptographic signature proving the message wasn’t modified?
DMARC: Does the email pass the policy the domain owner specified?

Failing authentication immediately raises the spam probability score — and as of 2025, results in outright rejection at several major providers. Gmail, Yahoo, Microsoft Outlook, and La Poste all now enforce DMARC for bulk senders, with non-compliant mail rejected at the SMTP level (not just filtered).

3. Content Analysis

Content signals are evaluated but weighted less heavily than reputation in modern systems:

Text patterns: Machine learning models scan for language associated with spam
HTML structure: Broken HTML, excessive inline styles, or image-only emails raise risk
Link reputation: URLs in your email are checked against blocklists and phishing databases
Text-to-image ratio: Image-only emails hide content from text analysis — treated as suspicious
Header analysis: Missing or malformed email headers raise flags

4. Engagement Signals

Gmail and other ISPs use their users’ behavior as a spam signal:

If users at that ISP frequently open emails from your domain → strong inbox signal
If users rarely open or delete without reading → weak inbox signal
If users move emails from inbox to spam → strong spam signal
If users move emails from spam to inbox → strong inbox signal

This is why engagement-based segmentation (sending only to active openers) isn’t just about open rate optimization — it directly shapes how ISPs route your future mail.

How Gmail’s Spam Filter Works

Gmail uses a multi-layer filtering system:

Compliance check: Does the sender meet bulk sender requirements (authentication, one-click unsubscribe, TLS)? As of November 2025, non-compliant bulk mail is rejected at the SMTP level with permanent 5xx error codes — the email never enters Gmail’s infrastructure.
IP reputation check: Is the sending IP known for spam?
Authentication verification: Does SPF/DKIM/DMARC pass?
Content classification: Machine learning analysis of message content
Engagement lookup: How do Gmail users typically interact with this sender?
Policy enforcement: Does the sender’s DMARC policy require action?

All of these are evaluated, and the result determines whether the message is rejected outright, goes to inbox, promotions, or spam. The shift to SMTP-level rejection in November 2025 means non-compliant mail no longer reaches filtering — it’s refused at the connection layer.

How Microsoft Outlook’s Filter Works

Microsoft began enforcing bulk sender requirements on May 5, 2025 — adding 500M+ Outlook, Hotmail, and Live.com mailboxes to the list of providers that actively reject non-compliant mail.

Non-compliant bulk mail to Microsoft-hosted addresses now returns 550 5.7.515 — a permanent rejection. Microsoft’s enforcement focuses on:

Authentication: SPF, DKIM, and DMARC required for bulk senders
One-click unsubscribe: RFC 8058 List-Unsubscribe headers required
Valid reverse DNS: PTR records must resolve correctly
Low complaint rate: Spam complaints above threshold trigger filtering

Microsoft’s SmartScreen filter additionally evaluates IP reputation (heavily weighted) and content patterns. Unlike Gmail, Microsoft places less weight on engagement history and more weight on IP-level signals — which means shared IP reputation issues affect Outlook deliverability more than they affect Gmail.

How Blocklists Interact with Spam Filters

ISPs consult external blocklist databases (DNSBLs) as part of their filtering decision:

Spamhaus: Most widely used blocklist; listing here causes significant delivery problems across all ISPs
Barracuda: Used heavily by Outlook and enterprise mail systems
SURBL: URL-based blocklist checking links in email content
URIBL: Another URL-based blocklist

Being listed doesn’t automatically block all your mail — ISPs use blocklists as one signal among many — but listing on Spamhaus or similar major lists can drop delivery rates by 30–50%.

Why the Same Email Performs Differently at Different ISPs

Your reputation is separate at each ISP. A domain that Gmail views as “High” may be “Medium” at Outlook and “Low” at Yahoo because:

Each ISP’s users have different complaint rates for your mail
Your IP reputation differs in each system’s database
Each ISP weights signals differently

This is why checking Google Postmaster Tools alone isn’t sufficient — you need visibility across all major ISPs simultaneously. InboxEagle monitors Google Postmaster, Yahoo Sender Hub, and seed list placement across 20+ providers in real time.

What Triggers a Spam Filter

The highest-risk actions that trigger spam filtering:

High spam complaint rate: Above 0.10% at Gmail, above 0.30% for Yahoo
Sending to spam traps: Addresses that were never valid or were abandoned years ago
Authentication failures: SPF/DKIM/DMARC not passing
Sudden volume spikes: 10x normal volume in a single send
Very low engagement: High send volume with very few opens/clicks signals poor list quality
Blocklist listing: IP or domain appearing on major blocklists

The Bottom Line

Spam filters aren’t keyword matchers — they’re multi-signal reputation engines that evaluate your sending history, authentication, and audience behavior simultaneously. Trying to “trick” them with content tweaks while ignoring the underlying reputation signals doesn’t work and hasn’t worked for over a decade.

Sender reputation is the primary signal — complaint rate, bounce rate, and engagement history outweigh content signals at every major ISP
Authentication is table stakes — SPF, DKIM, and DMARC failures cause immediate, measurable rejection across Gmail, Yahoo, Outlook, and La Poste
Engagement matters per-provider — your reputation at Gmail is separate from your reputation at Outlook; you can be filtering at one and clean at the other
Blocklist listing drops delivery by 30–50% — check before major sends; one blocklisted IP can compromise an entire campaign
ISPs respond to patterns, not single sends — a complaint rate spike from one campaign affects inbox placement for the next 30–60 days

Understanding how spam filters work helps you build an email program that works with ISP filtering systems rather than against them. For a look at how AI takes this further — continuously analyzing all these signals across sends — see How AI Monitors Email Deliverability: What’s Actually Different. InboxEagle monitors all the signals ISPs evaluate — reputation, authentication, placement — and alerts you when anything changes. Start a free trial →