Click Fraud Prevention in LLM Interfaces: Cryptographic Attribution

In conversational ad networks, click fraud can drain advertiser budgets within minutes if automated scripts replay tracking tokens. Protecting advertiser capital is paramount.

Multi-Layer Verification Pipeline

Every ad impression generated by MatrixAds returns a cryptographically signed HMAC token containing an expiring timestamp and unique auction nonce:

const clickToken = generateClickToken({
  adId: winner.ad.id,
  publisherId,
  clearingPriceMicros: clearingPriceMicros.toString(),
  auctionId: crypto.randomUUID()
});

Two-Phase Impression Proof & Redis Lua Budget Locks

When a click occurs, our /api/v1/track/click handler executes 4 anti-fraud checks prior to double-entry financial settlement:

• HMAC & Expiry Verification: Token age must be < 2 hours (7,200s).
• Datacenter IP Trap: Blocks AWS, GCP, Azure, DigitalOcean, and Tor proxy IPs.
• Two-Phase Impression Proof: Requires prior impression_seen:${auctionId} token in Redis from SDK viewport observer.
• Atomic Redis Budget Reservation: Executes atomic decrement before Prisma double-entry transaction to prevent negative balance race conditions.

If fraud is detected, the click is logged as invalid with an explicit fraudReason (e.g. DATACENTER_PROXY_BLOCKED, FRAUD_UNVERIFIED_IMPRESSION, SUBNET_VELOCITY_TRAP) without debiting the advertiser account, while ensuring the end user is still redirected via 302 to preserve user experience.