For years the rule was simple: a human shops, a bot scrapes. That is changing. AI assistants from large tech companies can now compare products, fill carts and in some cases check out on a customer's behalf. This holiday season is the first in which many stores will see this traffic in meaningful amounts.
That creates a new problem. Your fraud tools were built to block automation, and now some automation is a paying customer.
What agent traffic looks like
A shopping agent usually runs in the cloud, not on the customer's phone. So its requests come from a datacenter network, just like a scraper's. It moves faster than a person and may skip the pages a human would look at. On the surface it looks a lot like the bots you have been blocking.
How to tell them apart
Who runs the network. Legitimate agents run on the infrastructure of the companies that make them and usually say who they are in their user agent. A request claiming to be a well-known assistant but coming from a cheap hosting provider or a proxy is lying.
What it is doing. An agent buying one item for one customer behaves very differently from a script trying fifty cards or buying every unit of a limited product.
Whose card it is. A real agent pays with its customer's card, which usually matches that customer's country. A run of mismatched or prepaid cards points to fraud, however polite the user agent.
A practical policy for this season
- Keep blocking datacenter traffic on sensitive actions by default.
- Make exceptions for known agent providers you have chosen to allow, checked by network, not just by user agent.
- Keep your normal per-network limits on cards, accounts and limited stock for everyone, agent or not.
Fraudex tells you the network and organisation behind every IP, so an exception for one provider is a single rule, not a guess based on a header anyone can fake.