What Did Pichai Say About AI Agents and Websites?
That agents will reach websites in two ways, and both will grow. According to Search Engine Journal’s October 22, 2026 report of his interview with Alex Heath on the Sources podcast, Pichai described protocols built for agents as “a far more scalable approach” and agents that click through pages in a browser as “a very elegant way to work too, because that’s what you as a user would do.” He added: “I think you have both ways and they’re going to develop” (Search Engine Journal).
He also called this “a complex moment,” said “there’s going to be a chaotic evolution there,” and spoke of “working towards more interoperable standards.” Asked how Google protects an ads business that depends on engagement when agents do the browsing, he answered: “The right answer longer term is you put users first and say what would users want.” Search Engine Journal notes the answer did not address how Google’s ads would work when an agent browses. It also reports that Google plans to make Gemini’s agent features, currently available mainly to Ultra and Pro subscribers, more widely available “over the next few months” (Search Engine Journal).
A second Search Engine Journal report of the same interview quotes Pichai on Google’s products: “Our vision in the near future is it’s one box,” and “the agentic layer offers us a chance to bring a unified product experience better than ever before.” He said the share of queries “which are more agentic, has been steadily increasing over time” (Search Engine Journal). That report raises the question of whether referral traffic from people survives, but as the author’s question, not as anything Pichai said.
Every Pichai quotation here is as Search Engine Journal published it. The interview itself is on YouTube; we did not check the quotes against its audio, and neither report gives the recording date.
What Is Google-Agent?
Google-Agent is the name Google uses for agents running on its own infrastructure when they visit a page and act on a person’s request. Google’s documentation defines it as “used by agents hosted on Google infrastructure to navigate the web and perform actions upon user request,” and lists mobile and desktop variants whose user agent strings contain the token “Google-Agent” (Google Search Central).
It sits in Google’s list of user-triggered fetchers, alongside tools such as Feedfetcher, Google Read Aloud, Google Site Verifier and Gemini Notebook. A user-triggered fetcher differs from a crawler such as Googlebot: a crawler visits pages on Google’s own schedule to build an index, while a user-triggered fetcher visits because a person asked a Google product to fetch something (Google Search Central). The page was last updated August 19, 2026, when we read it.
Does robots.txt Stop Google-Agent?
Google’s documentation says not reliably. The user-triggered fetchers page states: “Because the fetch was requested by a user, these fetchers generally ignore robots.txt rules” (Google Search Central). The page does not give Google-Agent a robots.txt token for opting out, as Google does for some of its crawlers. We therefore treat robots.txt as an unreliable control for this fetcher, and we make no claim about how Google-Agent behaves on any particular site.
| Question | What Google’s documentation says | Source |
|---|---|---|
| What is Google-Agent? | “used by agents hosted on Google infrastructure to navigate the web and perform actions upon user request” | User-triggered fetchers |
| Does it follow robots.txt? | User-triggered fetchers “generally ignore robots.txt rules” | User-triggered fetchers |
| Can the user agent be trusted? | No: “The user agent string can be spoofed,” so verify requests | User-triggered fetchers |
| How do I verify a request? | Check the source IP against the published ranges; agents are listed in user-triggered-agents.json | User-triggered fetchers |
| Is there an authentication option? | Google is experimenting with Web Bot Auth for Google-Agent, identity agent.bot.goog | User-triggered fetchers |
This is a different question from whether robots.txt governs how content already crawled is used for AI. Google’s Gary Illyes addressed that in an October 21, 2026 Search Central video, calling robots.txt an access-preference protocol that does not dictate what a system does with data after crawling; our guide on what Google said robots.txt can and cannot do about AI crawlers covers it, and whether blocking a crawler removes you from search covers the search side.
What Are the Protocols Pichai Mentioned?
Ways for agents to deal with a site directly, without reading its pages like a person. Search Engine Journal quotes Pichai citing “a lot of momentum behind MCP or agent protocols or payment protocols.” MCP is the Model Context Protocol, an open standard for connecting AI tools to outside services. The report adds that UCP, the Universal Commerce Protocol, is an open standard for agent commerce that Google built with a group of companies, and that Google’s UCP page says merchants can adopt it to enable agent actions in AI Mode and Gemini, starting with direct buying (Search Engine Journal).
UCP is aimed at merchants selling products, not at publishers or company newsrooms. We did not read the UCP documentation and do not assess it here.
How Can a Site Tell When an Agent Visits?
By the user agent token, checked against Google’s published IP ranges. Requests from Google-Agent carry the “Google-Agent” token, but because the string can be spoofed, Google points site owners to its IP range files, and to Web Bot Auth for Google-Agent while that is being tested (Google Search Central). Our guide to verifying that a crawler is what it claims to be walks through the reverse-DNS and IP checks.
Google publishes no figure for how often Google-Agent visits sites, and we found none elsewhere that we would rely on. Our earlier look at whether AI agents fill out your contact form reached the same conclusion: nobody has published a trustworthy measure of agent traffic to company websites.
What This Means for Anyone Buying a Placement
If agents read pages on a person’s behalf, a press page, newsroom post or placed article may be fetched by software that robots.txt does not reliably govern. Our recommendations, none of them a Google rule:
- Do not rely on robots.txt for anything sensitive. Put embargoed or private material behind real access controls, such as a login.
- Make the page legible to a reader and to software. A placement a person can read quickly, with names, dates and facts stated plainly, is also easier for an agent to summarize accurately.
- Count agent visits carefully. Look for the Google-Agent token in server logs, and verify against Google’s IP ranges before counting a visit. See how to tell which AI sent a visitor.
- Do not buy on a promise. A paid placement is advertising, labeled as such, and nothing about agentic search means payment buys coverage, rankings, AI citations or permanence.
Methodology and Limitations
Four limits apply. First, every Pichai quotation reaches us through Search Engine Journal’s two reports; we did not check them against the interview recording, and the reports do not give its date. Second, Google’s robots.txt statement covers user-triggered fetchers as a group; the documentation does not give Google-Agent a separate rule. Third, we did not test Google-Agent against any site. Fourth, Google’s agent features are, per Search Engine Journal, mainly available to Ultra and Pro subscribers today, so real-world agent traffic may still be small. There is no chart, because the sources publish no comparable figures.
Conclusion
Google’s own documentation tells site owners that a text file will not reliably keep a user’s agent away, and Google’s chief executive expects agents to do more browsing, not less. The uncomfortable conclusion for a buyer is that a public page can be read by software acting for anyone, and the dependable controls are access controls and a page worth reading.
Frequently Asked Questions
What is Google-Agent?
Google describes it as a user agent “used by agents hosted on Google infrastructure to navigate the web and perform actions upon user request,” and lists it among its user-triggered fetchers (Google Search Central).
Does robots.txt block Google-Agent?
Not reliably. Google says its user-triggered fetchers, the group that includes Google-Agent, “generally ignore robots.txt rules” because a user requested the fetch (Google Search Central).
How can I tell a real Google-Agent request from a fake one?
Google says the user agent string can be spoofed. Check the request’s IP address against the ranges Google publishes in user-triggered-agents.json; Google is also testing Web Bot Auth for Google-Agent (Google Search Central).
What did Sundar Pichai say about AI agents and websites?
Search Engine Journal reported on October 22, 2026 that Pichai told Alex Heath agents will reach sites both by browsing and through protocols, calling protocols “a far more scalable approach” and saying “you have both ways and they’re going to develop.”
Will agentic search send traffic to my website?
The sources we read do not say. Search Engine Journal raised the question of referral traffic, but neither of its reports attributes an answer on it to Pichai.
Can a paid placement make AI agents cite a company?
No source we read says so. A paid placement is advertising that should be labeled, and payment does not buy AI citations.
Sources and Methodology
This Explainer is built from Google’s user-triggered fetchers documentation and Search Engine Journal’s two October 22, 2026 reports of Sundar Pichai’s interview, all read on October 23, 2026. Recommendations in the buyer section are house advice, not Google rules. We did not test Google-Agent or check the quotations against the interview recording.
- platform documentation Google Search Central, Google’s user-triggered fetchers (last updated August 19, 2026). Google-Agent definition, user agent token, robots.txt sentence, spoofing warning, IP range files and Web Bot Auth. Read October 23, 2026.
- trade press Search Engine Journal (Matt G. Southern), Pichai Says AI Agents Will Click Through Pages And Use Protocols, October 22, 2026. Pichai quotations on browsing, protocols and ads; UCP; Gemini agent availability. Read October 23, 2026.
- trade press Search Engine Journal (Roger Montti), Google CEO Sundar Pichai Views Search Absorbed Into Agentic Layer, October 22, 2026. “One box” and agentic-layer quotations. Read October 23, 2026.
- official Sources podcast, interview with Sundar Pichai by Alex Heath, on YouTube. Linked by Search Engine Journal; listed for reference, quotations not checked against it. Read October 23, 2026.