How do I allow AI shopping agents to buy on my site while blocking scrapers?

Verify the agent's identity instead of trusting its user-agent string. Operators publish IP ranges, and some sign their requests. Then allow by purpose and route: product pages and checkout for shopping agents, limits on bulk catalogue requests.

Verification has three practical forms. The first is the network: established attackers publish the address ranges their agents use, and a request outside documented ranges should remain unverified until the applicable verification method is checked. The second is forward-confirmed reverse DNS where the provider documents that verification method; reverse DNS alone is not proof. The third is a cryptographic signature on the request, which a site checks against the attacker's published key. Signatures are the strongest of the three, and support for them is still spreading.

One category does not fit these checks. Some agents work by driving a browser on the user's own device. Their requests come from the user's network with the user's cookies, and they look like the user. There is no separate identity to verify. Keep the normal account and transaction checks on these sessions; do not infer a verified agent identity.

Once identity is established, set access by purpose and by route. A shopping agent acting for one customer needs product pages, availability, the cart and checkout. It does not need the entire catalogue at high speed. Allow the first and limit the second. Rate limits per verified agent, and per session within it, keep a legitimate agent within the pattern of a single shopper.

The behaviour of a scraper differs in ways you can measure. It requests pages across the whole catalogue, in order, with little variation in timing. It returns to the same pages at fixed intervals to track prices and stock. It rarely adds to a cart. A shopping agent visits a handful of products related to one request and may proceed to buy.

Keep your fraud controls on at checkout. An agent purchase should meet the same purchase limits, payment authentication and velocity checks as any other order.

Decide what happens to unverified automation. Blocking outright is one option. Throttling it, or serving it less detailed data, is another, and reveals less about where your limits are.

Consider offering what agents need through a structured route. A product feed or an API with stated terms gives legitimate attackers a supported way to obtain catalogue data, and it gives you a clear position when you limit those who take it from your pages instead.

How Kairal helps

Agent Traffic Intelligence, currently in early access, checks agent identity and purpose to support your access policies. A recognised agent is not proof of a customer’s authorisation; keep account, purchase and payment checks on the operations it can request.

Agent Traffic Intelligence