Google tells sites how to let its AI agent through bot filters
Six causes can stop the AI agent of Google, also when a site lets it in.
On 17 November 2026, Google released a document for owners of sites that stop bots with firewalls and filters. Google-Agent, its AI agent, works for a person and can find items, add them to a cart and make bookings. It tells owners to let correct requests go without CAPTCHAs, tests that tell a person from a bot.
How it works
All requests have a header that contains the name Google-Agent. They also come from an address on the lists that Google gives. The lists had 20 address blocks on 17 November.
The proof
The document gives six causes that can stop correct requests. One cause is a limit for each address, because many users send requests from the same pool of addresses. Google tells that a header name is not proof, because a person can write it.
What is not known
Google tells that only some requests have a signature, which is a digital proof of the sender. Sites must also check the address lists, and Google makes new lists each day. The document gives no count of sites that stop the agent.
Insight: Google gives sites the work of letting its agent in
The document shows that the agent usually ignores the rules that tell bots to not open some parts of a site. It tells sites to remove CAPTCHAs for it. On 2 November, Google also released an Ads server with four write tools on by default. This reads as a move to open access by default, with the cost for the site owner. Leaders must select who lets agents in.
Sources
Posted