Google outlines two ways AI agents may access websites
Google’s Sundar Pichai said AI agents are likely to reach websites in two ways, by browsing like people and through dedicated protocols.
Google CEO Sundar Pichai said he expects AI agents to reach website content in two main ways, by browsing the web in a human-like way and by using protocols built for agents. According to Search Engine Journal, he said both approaches will grow, while protocol-based access may scale better over time but take longer to mature.
The comments came during an interview on the Sources podcast. The report also notes that Google is backing the Universal Commerce Protocol, or UCP, and that Google lists Google-Agent as the user agent used when hosted agents browse the web.
What changed
Pichai described a split path for how AI agents may interact with websites. One path is browser-based, where an agent moves through pages much like a person would. The other is protocol-based, where websites and services connect through standards designed for agent use.
He said both methods are likely to develop at the same time. He also said protocols are a more scalable option, but they will take longer to build out. In the meantime, browser-style access remains part of the picture. The report adds that Google-Agent is used by agents on Google’s servers to navigate the web and take actions requested by users.
What it means for your website
For business owners, this suggests that regular website pages still matter for AI visibility. If agents continue to browse websites the way people do, then the content and flows already available on your site may still be used while broader standards are worked out.
It also means technical controls may not work the same way you expect. Search Engine Journal reports that Google classifies Google-Agent as a user triggered fetcher, and that these fetchers usually do not follow robots.txt directives because the request comes from an end user. If your site depends on robots.txt alone to manage access, that may not fully cover agent activity.
What to do now
- Keep key pages accessible and functional for normal browser use, especially pages involved in research, booking, shopping, or lead capture.
- Review server logs for traffic from Google-Agent and compare it with Google’s published IP ranges when investigating agent activity.
- Do not assume robots.txt will block all agent visits, especially when the visit is triggered by a user request.
- Monitor Google’s work on UCP and related agent protocols if your business relies on online transactions, since Google says merchants can use UCP for agentic actions in AI Mode and Gemini.
- Watch for broader rollout of Gemini agent features over the next few months, since wider availability could increase how often agents interact with business websites.
More from SEO News
All SEO News-
· Search Engine Journal
United Allows Search Indexing but Restricts AI Agents
-
· Search Engine Journal
Google details search referrer values for analytics and logs
-
· Search Engine Journal
Google Sets Gemini 4 Hydrogen as AI Mode Default Worldwide
-
· Google Search Status Dashboard
Google begins rolling out the November 2026 core update