Google has just introduced a new user-agent called "Google-Agent", designed to allow its AI agents to crawl the web and perform actions at users' requests. A strong signal of the rise of autonomous agents within the Google ecosystem.
Key takeaways:
- Google added a new user-agent called "Google-Agent" in the "user-triggered fetchers" category.
- It is used by agents hosted on Google’s infrastructure to browse the web and perform actions at users' request.
- Project Mariner is cited as a concrete example of use.
- Google is also experimenting with the web-bot-auth protocol using the identity https://agent.bot.goog.
Since March 23, 2026, Google has officially documented the arrival of a new user-agent: Google-Agent. It joins the list of "user-triggered fetchers", meaning robots triggered not by the search engine itself but by a direct action from a user.
Concretely, when you ask a Google agent to perform a task on the web, this is the user-agent that presents itself to the servers of the visited sites. Google specifies that the rollout is happening gradually over several weeks.
A user-agent tied to Google’s AI agents
The Google-Agent user-agent is associated with agents hosted directly on Google's infrastructure. These agents can navigate web pages and perform actions on them, much like a human user but in an automated way.
Google cites Project Mariner as an example of use. As a reminder, Project Mariner is Google's AI agent capable of taking control of a browser to carry out tasks on behalf of the user: filling out forms, performing searches, interacting with web interfaces, etc.
On the technical side, Google-Agent uses the IP ranges referenced in the file user-initiated-agents.json, which allows site administrators and developers to identify and filter these requests if necessary.
The web-bot-auth protocol takes the stage
Alongside this new user-agent, Google says it is experimenting with the protocol web-bot-auth, an emerging standard intended to authenticate bots on the web. For this, Google uses the identity https://agent.bot.goog/.
This protocol aims to bring more transparency to interactions between automated agents and websites: rather than a simple user-agent that anyone can spoof, web-bot-auth would allow verification that a bot is indeed who it claims to be. This is an important issue for publishers who want to control access to their content as AI agents proliferate.
The article "Google launches a new user-triggered fetcher for AI" has been published on the site Abondance.
