What signals do AI models analyze (content, engagement, headers, volume, etc.)?

AI filtering models analyze multiple signal categories: content features (text patterns, HTML structure, link characteristics), header information (authentication results, routing patterns, formatting), engagement data (open rates, complaint rates, interaction patterns), and volume metrics (sending velocity, burst patterns, consistency).

Content signals include word patterns, phrase combinations, image to text ratios, and structural elements common in spam. Header analysis examines authentication pass/fail, header malformation, and routing anomalies. Engagement signals reveal whether recipients want the mail.

The models combine these signals with varying weights learned from training data. No single signal determines filtering; the combination of many weak signals creates strong classification. This multi dimensional analysis makes gaming the system difficult since optimizing one dimension while neglecting others still triggers filtering.