How to detect Google's AI bots

You detect Google's AI bots by matching the token each one declares (Google-Agent, Google-GeminiNotebook or GoogleOther) and confirming the IP is in Google's published range files or passes a reverse and forward DNS check. Google-Extended is different because it is only a robots.txt control over how Google may use pages Googlebot already crawls, so it never sends a request. Verisoul verifies the fetchers against Google's network and returns them as agent id google_agent, googleother or google_notebooklm.

Detection method

Self-declaredThe agent names itself in its user agent, and Verisoul confirms it with network, device or behavioral data. A user agent alone is never enough.

Signals that identify it

NetworkSource IP confirmed as the operator's own network

Google documents its AI-related crawlers and fetchers across common crawlers and user-triggered fetchers. GoogleOther joined the list on Apr 20, 2023 for one-off research crawls, Google-NotebookLM arrived in Oct 2025 and was renamed Google-GeminiNotebook on Jul 17, 2026, and Google-Agent was added on Mar 20, 2026 as the identity for agents hosted on Google infrastructure that browse and act on pages.

Google is also the biggest crawler operator overall. Googlebot, which feeds Google's AI features as well as Search, made up about 24.5% of AI-bot traffic in Cloudflare Radar data for July 2026, and Cloudflare said Googlebot was over 28% of the verified-bot traffic it saw in 2025. Radar does not break out Google-Agent or the Gemini fetchers.

On sites that use Verisoul, from Oct 29 to Nov 1, 2026 (Nov 1 partial), the largest named Google source was Google Read Aloud, the fetcher behind Google's text-to-speech reading of pages, which produced about twice as many sessions as Meta's Muse. Google Agent was much smaller at 1% of the personal-agent sessions we named, though it is the one that fills forms, which is why fraud teams ask about it.

How Google's AI bots show up on your site

Every fetcher runs from Google-owned IPs, and reverse DNS for user-triggered fetchers resolves to gae.googleusercontent.com or google-proxy hosts. Google-Agent sends a Chrome-based user agent for Linux desktop or Android with a Google-Agent token, and Google publishes its IPs in a dedicated file, user-triggered-agents.json. Google-Extended has no user agent at all, since it only tells Google whether pages Googlebot crawls may be used for Gemini training and grounding.

Check	What Google's AI bots do
User agent	Declared tokens Google-Agent, Google-GeminiNotebook, GoogleOther
Web Bot Auth signature	Experimental for Google-Agent only, as agent.bot.goog
IP ranges	Published JSON files plus reverse and forward DNS
robots.txt	GoogleOther follows it, user-triggered fetchers generally don't

Why Google's AI bots are easy to detect

Every documented fetcher declares a token and comes from Google IPs that pass DNS checks, so verification is straightforward, and Google-Agent is starting to sign its requests, which will make it stronger still. The confusing part is Google-Extended, which looks like a bot name but is only a robots.txt rule, so you can't block it at the firewall or find it in logs, and Google says it doesn't affect Search ranking.

The other trap is the NotebookLM rename, which broke allow and block rules that hard-coded the old token. Clicks from the Gemini iOS app can also arrive as an in-app browser carrying a real person, which should not be treated as a bot.

How Verisoul detects Google's AI bots

These fetchers are self-declared, so Verisoul reads the token and confirms the source IP is Google's own before naming it. For Google Agent we also look at the session itself, where a cloud-hosted browser with a virtual machine device profile and automated clicks, typing and page flow back up the network check.

API response

"agent": {
  "id": "google_agent",
  "name": "Google Agent",
  "provider": "Google",
  "type": "ai_agent",
  "detection_method": "self_declared",
  "detection_signals": [
    "network",
    "behavioral"
  ]
}

id is google_agent here, with googleother and google_notebooklm for the read-only fetchers
type is ai_agent, because Google Agent acts on pages for a person rather than only reading them
detection_method is self_declared, since the agent named itself in its user agent
detection_signals lists network (Google's own IP), device (a cloud VM browser profile) and behavioral (automated page flow)

See the Google Agent entry in Agent Atlas and the AI agent detection docs for every Google id.

01Person asks a Google agent for help
02Google-Agent opens your site from Google IPs
03It fills forms with automated input
04Your page sees a cloud VM browser

How Google's AI bots reaches your site

How to detect Google's AI bots step by step

Turn on Verisoul and read the agent object in the API response
Write separate rules for google_agent, which acts, and googleother and google_notebooklm, which only read
Use robots.txt with User-agent: Google-Extended to opt out of Gemini training and grounding
Update any rule that still names Google-NotebookLM

Should you block Google's AI bots?

GoogleOther, Gemini Notebook and Read Aloud only read pages, so they are a content decision, and Google-Extended belongs in robots.txt. Google Agent acts for a real person, so most teams let it browse and buy and send it to review at sign-up and payouts, where an automated session deserves a second look.

Questions about Google's AI bots

What is Google-Agent in my server logs?

It is the user agent for agents hosted on Google infrastructure browsing and taking actions for a person. Google added it on Mar 20, 2026 with its own IP file, user-triggered-agents.json.

Does blocking Google-Extended hurt my SEO?

No. Google says Google-Extended does not affect Search ranking, because it only controls whether pages Googlebot already crawls may be used for Gemini training and grounding. It never fetches pages, so it won't appear in your logs.

Gemini Notebook ignores robots.txt. How do I block it?

User-triggered fetchers like Google-GeminiNotebook generally ignore robots.txt because a person asked for the page. Block it with a firewall rule on the user agent plus Google's IP list, and update any rule still using the old Google-NotebookLM name.

Is GoogleOther the same as Googlebot?

No. GoogleOther is a generic crawler for Google teams' one-off research crawls, though it runs on Googlebot's infrastructure and obeys robots.txt.

How do I verify a request really comes from Google-Agent?

Check that the IP is in user-triggered-agents.json or passes reverse then forward DNS to a Google domain. Once Google moves past its experiment, you can also verify a Web Bot Auth signature pointing to agent.bot.goog.

Google's AI bots