In 2014, retired General Michael Hayden — former director of both the NSA and CIA — said something remarkable in public: "We kill people based on metadata." The comment, made at a Johns Hopkins event, was a candid acknowledgment of how intelligence agencies actually operate. Content analysis is labor-intensive. Metadata analysis is automated, scalable, and often more revealing than the messages themselves.
What Metadata Contains
Metadata is data about data. For a phone call, metadata includes: the two phone numbers involved, the timestamp of the call, the call duration, the cell towers used (revealing location), and the frequency of calls between those numbers. For a text message or chat: sender, recipient, time, message length (revealing something about content), device type, IP addresses, and whether the message was read. For a photo sent in a chat: the device it was taken on, GPS coordinates embedded in the file, the camera model, and timestamp of capture.
Individually, these data points seem innocuous. Aggregated over time and correlated across sources, they construct an extraordinarily detailed behavioral portrait. The pattern of who you communicate with and when can reveal your political associations, your medical situation, your romantic life, and your daily routine — without a single word of message content being analyzed.
How Platforms Collect Metadata
Every major messaging platform collects metadata. Signal, which has the strongest privacy posture of any widely-used messaging app, still collects your phone number, account creation date, and last-seen date. WhatsApp collects all of the above plus contact lists, usage frequency, device identifiers, and IP addresses. Facebook Messenger collects everything WhatsApp does and correlates it with your broader Facebook activity profile.
Government surveillance programs revealed by Edward Snowden in 2013 operated primarily through metadata: the NSA's PRISM and Upstream programs collected metadata at scale from major US tech companies under Section 702 of FISA. The legal threshold for accessing metadata has historically been lower than for content — further incentivizing its collection by surveillance agencies.
Platforms That Generate No Metadata
The only way to eliminate metadata exposure is to build platforms that generate none worth collecting. An anonymous chat platform with no accounts has no "sender" identifier. Conversations transmitted over ephemeral WebSocket connections and never stored create no server-side metadata record of who talked to whom. Temporary usernames that exist for a single session produce no longitudinal data. When there is no persistent identity, there is no metadata to aggregate — because there is no entity across sessions to aggregate it about.