Full Answer

Meta designed CAPI to receive hashed customer identifiers — email, phone, IP — and match them against its user graph to attribute conversions. The system assumes every incoming event represents a real person completing a real purchase. There's no agent_origin parameter, no bot flag, and no mechanism to separate a human clicking 'Buy Now' from a ChatGPT shopping agent completing the same transaction via WooCommerce MCP.

That blind spot creates a compounding problem. Meta's ad algorithm learns from your conversion data to find lookalike audiences. If 15% of your purchases come from AI agents, the algorithm builds audience models that include agent-like behavioral patterns — high cart velocity, no browsing history, zero page dwell time. The result isn't a dramatic failure; it's a slow, invisible erosion of targeting quality that's nearly impossible to diagnose from inside Meta's reporting.

The fix has to happen before the CAPI event fires. Server-side tracking that intercepts the purchase at the PHP hook level can tag agent orders based on behavioral signals like user-agent strings and session patterns, then either exclude them from the CAPI payload or pass them with a reduced value. This preserves the signal quality that delivers CAPI's measured cost-per-result advantage. For more on maintaining Meta's event match quality score with clean data, see How do I improve my Meta EMQ score on WooCommerce?.