Full Answer

When a WooCommerce store processes an AI agent order through MCP, the same PHP hooks that fire for a human checkout send purchase events to Google Ads. Smart Bidding doesn't see the difference — it ingests the conversion, notes the signal profile, and adjusts bids to find more users who look like that buyer. The problem is that agent-originated sessions don't share the behavioural patterns of human shoppers: they convert faster, skip browse pages, and carry no meaningful audience attributes for Google to target.

As the agent share of orders grows, the bidding model trains on an increasingly artificial signal. CPAs rise because the algorithm chases a profile that doesn't exist in any ad auction. Retargeting audiences fill with agent sessions that will never return through a paid click. Attribution reports show strong ROAS from conversions your campaigns didn't influence, masking declining real performance.

The fix sits at the event pipeline, not the ad platform. Tagging agent versus human at the server — before the conversion event leaves your infrastructure — keeps Smart Bidding training exclusively on human purchase signals. Without that separation, every agent order that shares a conversion endpoint with your ads actively degrades the model that allocates your budget. For a deeper look at how AI traffic distorts bidding signals, see AI Traffic Is Breaking Your Smart Bidding Signal.