Sybil Detection Through Timing Analysis - What Content Analysis Misses
Sybil Detection Through Timing Analysis
Content analysis for detecting bots is a losing battle. AI-generated text is good enough now that you cannot reliably distinguish it from human-written content. But timing? Timing is much harder to fake.
What Timing Reveals
Bots have distinct timestamp patterns that content analysis completely misses:
- Burst patterns - a sybil network posts 50 replies across different threads within a 2-minute window. Humans do not do this.
- Consistent intervals - posts exactly 45 seconds apart, every time. Humans have variable timing.
- Time zone inconsistency - an account claims to be in London but posts at 4am GMT consistently while being active during US business hours.
- Response latency - replying to a complex post within 8 seconds. Even fast readers need more time.
Why Content Analysis Fails
Modern language models generate text that is contextually appropriate, varied in style, and free of the obvious markers (repeated phrases, unnatural grammar) that early bot detection relied on. You can still catch lazy bots, but a well-prompted AI produces content that passes human review.
Timing is different because it requires operational discipline that most bot operators do not implement. They focus on making the content look real and forget that the metadata tells a story too.
Building a Timing Detector
The basic approach:
- Log timestamps for every action by every account
- Calculate inter-action intervals and look for suspiciously consistent patterns
- Cluster accounts that post in synchronized bursts
- Flag accounts whose active hours do not match their claimed location
Even simple heuristics catch a surprising percentage of coordinated inauthentic behavior. The key insight is that timing analysis scales better than content analysis - it is fast, cheap, and does not require expensive model inference.
Combine timing with network analysis (which accounts interact with each other) for the strongest detection.
Fazm is an open source macOS AI agent. Open source on GitHub.