A Plan for Spam

Paul Graham’s essay A Plan for Spam argues that content-based Bayesian filtering offers the most effective approach to spam.

The Core Insight

“The Achilles heel of the spammers is their message.” While spammers can circumvent other barriers, they must ultimately deliver readable content that software can analyze.

The Methodology

Training: Analyze corpora of spam and legitimate email, calculating word occurrence frequencies
Classification: Score new emails using the most statistically interesting words
Results: “Less than 5 per 1000 spams missed, with 0 false positives”

Advantages Over Rules

Statistical analysis discovers non-obvious indicators
Measures actual probabilities rather than arbitrary scores
Evolves naturally with spam tactics

The Business Perspective

“Spammers are businessmen” who send mail because response rates justify minimal costs. If filtering forces diluted messages, response rates decline and spam becomes unprofitable.

My Takeaway

This essay pioneered modern spam filtering. The insight that attackers must deliver interpretable content applies far beyond email.