Full Answer

The confusion starts because most analytics tools label everything non-human as "bot traffic" and filter it out. That's a problem when some of that traffic is placing orders. Training crawlers — GPTBot, ClaudeBot, Bytespider — send identifiable user-agent headers, respect robots.txt directives, and never interact with your cart or checkout. They're indexing your content for AI model training, not shopping. You can block them with a single robots.txt rule and lose nothing but bandwidth.

Shopping agents are fundamentally different. They run full Chromium browser instances, carry standard Chrome user-agent strings, and execute JavaScript — including your tracking scripts, if those scripts fire client-side. They add products to carts, fill shipping forms, and complete payment. Because their browser fingerprint matches a real visitor, your analytics has no reliable way to separate them from humans without server-side event inspection.

Seresa's breakdown of the six types of AI visitors your analytics mixes together (https://seresa.io/blog/ai-data-readiness/stop-calling-it-ai-traffic-the-six-types-of-ai-visitors-your-analytics-is-mixing-together) explains how these categories collapse in GA4 and why the distinction between a crawler and an agent is the difference between a cost centre and a revenue channel.