Answer first: A Google-Agent line in a server log is a user-triggered fetch, once you have checked it is really Google. Google’s fetchers page says agents hosted on Google infrastructure use that token to navigate the web and perform actions when a user asks. The page does not say this token fills the Search index or selects a supporting link in AI Overviews or AI Mode. ChatGPT search answers are an OAI-SearchBot question. Perplexity search links are a PerplexityBot question. A verified Google-Agent hit does not answer either one. Recheck the pages below before you brief anyone from a log line. This article records them for a reader on 18 October 2026.

What is Google-Agent?

Google splits its crawlers and fetchers into three groups on the verification page: common crawlers such as Googlebot, special-case crawlers such as AdsBot, and user-triggered fetchers. Google-Agent is in the third group.

The fetchers page gives two user-agent strings. Chrome/W.X.Y.Z is Google’s own placeholder for the version, not a version number to match character for character. The token to read is Google-Agent, and the string points at Google’s crawling docs.

`` Mozilla/5.0 (Linux; Android 6.0.1; Nexus 5X Build/MMB29P) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/W.X.Y.Z Mobile Safari/537.36 (compatible; Google-Agent; +https://developers.google.com/crawling/docs/crawlers-fetchers/google-agent) ``

`` Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko; compatible; Google-Agent; +https://developers.google.com/crawling/docs/crawlers-fetchers/google-agent) Chrome/W.X.Y.Z Safari/537.36 ``

The first is the mobile agent. The second is the desktop agent. Google says this token uses the IP ranges in user-triggered-agents.json.

The same page lists other user-triggered tokens that are easy to mix up with this one. Gemini Notebook (Google-GeminiNotebook, and the older Google-NotebookLM string, which Google says it supports until August 2026) fetches URLs a person added as notebook sources. Google Read Aloud fetches a page to speak it. Google Messages builds a link preview. None of those strings is Google-Agent. The page also says the list is not exhaustive: it covers requestors that are likely to show up in logs and that Google has been asked about. A token that is missing from the list is not automatically this one.

Google’s crawling changelog records that the user agent was added on 20 March 2026. The reason on that entry is that agents on Google infrastructure would use it to navigate the web and perform actions on user request. The tense on that entry is the tense of March. The description to use now is the fetchers page, not the rollout sentence.

Does a Google-Agent visit count as an AI citation?

No. Three products people lump together under “AI search” document three different paths.

Google Search, including AI Overviews and AI Mode. Google’s AI features guide says a page has to be indexed, and eligible to be shown in Google Search with a snippet, before it can be a supporting link. There are no extra technical requirements and no special schema. Meeting the requirements still does not mean Google will crawl, index, or serve the page. The optimisation guide adds that the Search Console property has to be included in Search generative AI features. Google-Agent is not that eligibility check. A line in a host log is not a row in Search Console.

The same optimisation guide keeps “agentic experiences” in a later section. It describes AI agents as systems that can perform tasks for a person, such as booking or comparing products, and says browser agents may read a screenshot, the page structure, and the accessibility tree. It frames that work as optional: look at it if it is relevant and you have time. That is a different job from being quoted in an overview.

ChatGPT. OpenAI’s crawler page says OAI-SearchBot surfaces sites in ChatGPT search. Sites opted out of OAI-SearchBot are not shown in ChatGPT search answers, though they can still appear as navigational links. ChatGPT-User is the other fetch. When a person asks ChatGPT or a custom GPT a question, OpenAI may visit the page with that agent, and the same agent covers some GPT Actions. OpenAI says ChatGPT-User is not used for crawling the web in an automatic fashion, that robots.txt rules may not apply because a user started the fetch, and that it is not used to decide whether content may appear in Search. The four OpenAI tokens are already split in the citations guide. Google-Agent is not one of them.

Perplexity. Perplexity’s crawler page says PerplexityBot surfaces and links sites in Perplexity search, and that it is not the crawler used for foundation-model training. Perplexity-User is the user-action fetcher. When someone asks a question, Perplexity says it might visit a page to help answer and include a link. The same page says that fetcher generally ignores robots.txt, because a user requested the fetch. A Google-Agent log line does not tell you what Perplexity showed.

So a morning of log lines can contain three different events: a search crawler, a user-asked fetch, and something spoofing a famous user-agent string. Only the verified user-agent, plus the vendor’s own rules, tells you which event it was.

Does robots.txt stop Google-Agent?

The fetchers page says user-triggered fetchers generally ignore robots.txt rules, because a user requested the fetch. The verification page’s table for that same category, which includes the user-triggered-agents.json file, says these fetchers ignore robots.txt rules. Keep both words. “Generally” is the list page. “Ignore” is the verification table. Neither page prints a User-agent: Google-Agent group and calls it an off switch.

Googlebot sits in the other category. The verification page says common crawlers always respect robots.txt for automatic crawls. A Disallow aimed at Googlebot is a Search control. It is not, on these pages, described as a Google-Agent control. Google’s technical requirements say one condition of index eligibility is that Googlebot is not blocked. The AI features guide says a supporting link has to be indexed. Blocking Googlebot is a decision about that eligibility. It is a blunt one if all you wanted was fewer agent fetches.

The Google-Agent section does not mention noindex, nosnippet, or Google-Extended. Do not borrow those tokens from a different guide and tell a client they stop this fetcher. The pages recorded here do not say that.

A firewall can still challenge the request. That is a host rule, not a sentence on Google’s fetchers page. If the challenge returns a block page, the agent fetched the block page. Confirm that on the URL you care about before you rewrite robots.txt.

How do I check the request is really from Google?

Google says the user-agent string can be spoofed. The verification page gives the manual check for one address.

  1. Take the IP from your own log. Run a reverse DNS lookup with host.
  2. Confirm the name is googlebot.com, google.com, or googleusercontent.com.
  3. Run a forward lookup on that name.
  4. Confirm the forward lookup returns the same IP.

The worked examples on Google’s page are other clients. Do not paste those addresses into a report and label them Google-Agent. Use the IP that requested your URL.

Then match that IP to the published list for this token. The file is user-triggered-agents.json. Google says the addresses are in CIDR form. The copy fetched while this article was written carried a creationTime of 2026-10-16T14:46:09.000000. Fetch the file again on the day you check. Do not copy ranges out of a blog post, including this one. An old list is how a firewall allows the wrong addresses, or blocks a range Google has since published.

The fetchers page also prints reverse-DNS masks for user-triggered fetchers as a group: a gae.googleusercontent.com name, or a google-proxy- name under google.com, depending on whether the fetcher is user-owned or Google-owned. The Google-Agent paragraph itself does not assign this token to one of those two masks. Pass the DNS steps and the JSON list. Do not invent a hostname the paragraph does not print.

A string that says Google-Agent, from an IP that fails the DNS check and is absent from that file, is not this fetcher.

What does a missing Web Bot Auth signature mean?

Google is testing Web Bot Auth, a draft way for a bot to sign a request. The page says the test is experimental for two reasons: the specification is an IETF draft and may change, and IP-and-user-agent checks are still the method everything else is built around.

The practical limits are on the same page. Not all Google user agents use Web Bot Auth. Google is not yet signing every request from agents that use it. Google says to keep checking IP addresses, reverse DNS, and user-agent strings as well.

A subset of Google-Agent requests are signed. Google says those requests authenticate as https://agent.bot.goog. On a participating request, the Signature-Agent header is g="https://agent.bot.goog". The public keys are published at https://agent.bot.goog/.well-known/http-message-signatures-directory. A missing signature does not, on that page, mean the request is fake. Google says to fall back to the IP check because not every request is signed.

The allowlist steps are written for a developer or a system administrator, or for a host that already understands the protocol. A small firm does not need to verify HTTP message signatures in order to read one log line. The DNS check and the current JSON file are the check Google still tells you to keep. A signature is an extra. It is not a citation, and it is not a ranking.

What should the page contain if an agent might use it?

If someone asks an agent to book, compare, or request a quote, the agent has to be able to use the page you already publish. Google’s optimisation guide calls that preparation optional. For a service firm the useful version is ordinary: the answer in text, and a form a person can finish.

The quote form guide is that checklist. A real button, a label on each field, the phone number as text, and no banner sitting on Submit. This article does not repeat it. A verified Google-Agent fetch does not mean the form was submitted.

The sentence next to the form still has to be true. Name the service and the places you actually cover. A Douglas firm that only works on the Island should say so. A firm that covers three towns should name those towns, not a near-copy page for every parish. The optimisation guide says that making separate pages for query variations, mainly to manipulate rankings or generative AI responses, violates Google’s scaled content abuse policy. An agent that can click Submit cannot repair a coverage line that was never written.

That same sentence is what a supporting link needs as text. The AI features guide says important content should be available in textual form. Google-Agent does not replace that requirement, and a fetch does not put the sentence into an overview.

What should I check when the token appears in a log?

Do this on the host customers use. Write the date next to what you find.

  1. Read the user-agent. Confirm the token is Google-Agent, not Googlebot, Google-InspectionTool, Google-GeminiNotebook, or Google-Read-Aloud.
  2. Run the reverse and forward DNS checks on the IP from that line. Then match the IP to a fresh copy of user-triggered-agents.json. Record the file’s creationTime.
  3. If a Web Bot Auth signature is present, Google’s page says a signed Google-Agent request authenticates as https://agent.bot.goog. If the signature is absent, do not call the request fake on that basis alone.
  4. Leave robots.txt alone unless you have a reason that is written on the fetchers page. A Disallow for this token is not described there as an off switch.
  5. If the question is whether the URL can be a supporting link in AI Overviews or AI Mode, leave this log. Read the indexed page against the AI features guide, and read the Search Console include setting the optimisation guide names. This log does not show either one.
  6. If the question is whether an agent could send a quote, open the live form logged out and use the quote form guide. A fetch is not an enquiry.
  7. If the question is ChatGPT or Perplexity, look for those vendors’ own tokens. A Google-Agent line is the wrong evidence.

What should I not claim?

  • That a Google-Agent hit is an AI Overview, an AI Mode link, a ChatGPT citation, a Perplexity citation, a click, or an enquiry.
  • That the 20 March 2026 changelog line still means the token is rolling out over the next few weeks. Use the fetchers page for what it is now.
  • That a robots.txt block aimed at Google-Agent, Googlebot, or Google-Extended is documented as this fetcher’s off switch.
  • That a missing Web Bot Auth signature proves the request is spoofed, or that a present signature is a mention.
  • That IP ranges copied from this article will still be right next month. Fetch the JSON file.
  • That every Google request appears on the fetchers list. Google says the list is not exhaustive.
  • A result, a town, or a before-and-after Halo has not published with evidence. This guide does not add one.

Where Halo fits

Halo’s search, AEO and content service covers the public page: the answer in the first screen, internal links, and structured data that matches the visible words. Google-Agent sits in the host log. Halo does not need a server login to read the page an agent would have to use. Verifying a single IP is something the person who holds the log can do with the steps above.

To start, send the URL through the free AI audit. If a log line is why you wrote, paste the user-agent and the URL it requested, and say whether the IP matched the current JSON file. Halo does not guarantee a citation, an agent visit, an enquiry, or a change in anyone’s overview.

FAQ

Is Google-Agent the crawler behind AI Overviews?

No. Google’s fetchers page describes Google-Agent as a user-triggered agent that navigates the web and performs actions when a user asks. The AI features guide says a supporting link in AI Overviews or AI Mode has to be indexed and eligible for a snippet. A log line is not that eligibility, and it is not an impression.

Will blocking Googlebot in robots.txt stop Google-Agent?

The pages recorded here do not say that. The verification page says common crawlers such as Googlebot respect robots.txt for automatic crawls. Google-Agent is on the user-triggered list, and that list’s introduction says those fetchers generally ignore robots.txt. Treat the two tokens as different controls.

If there is no Web Bot Auth signature, is the request fake?

Not by that absence. Google says its Web Bot Auth test does not sign every request, and that you should fall back to IP addresses, reverse DNS, and the user-agent string. A signature on a Google-Agent request is documented as the https://agent.bot.goog identity. A missing signature is not a verdict.

Does a Google-Agent fetch mean an enquiry was sent?

No. Google describes navigation and actions a user requested. It does not describe a completed quote, a booking, or a lead. Whether a form can be used is a separate check, in the quote form guide. A verified fetch is a request in a log.

Is Google-Agent the same kind of visit as ChatGPT-User or Perplexity-User?

Each vendor documents a user-triggered fetch separately from its search crawler. OpenAI says ChatGPT-User is not used to decide whether content may appear in Search. Perplexity says that when someone asks a question, Perplexity-User might visit a page to help answer it and include a link. A Google-Agent line is not either token, and it does not show what those products displayed.

Does an Isle of Man firm use a different Google-Agent?

No. The fetchers page does not describe an Isle of Man token or a UK-only token. The same checks apply on a .im host and a .co.uk host. The coverage sentence on the page still has to name where you actually work.

Sources

  • Google crawling infrastructure, user-triggered fetchers, including the Google-Agent user-agent strings, the user-triggered-agents.json ranges, the note that user-triggered fetchers generally ignore robots.txt, neighbouring tokens, and the statement that the list is not exhaustive.
  • Google crawling infrastructure, verify requests, including the three crawler categories, the reverse and forward DNS checks, and the user-triggered row that includes user-triggered-agents.json.
  • Google crawling infrastructure, Web Bot Auth, including the experimental status, the statement that not every request is signed, and the https://agent.bot.goog identity for signed Google-Agent requests.
  • Google crawling infrastructure, changelog, 20 March 2026 entry adding the Google-Agent user agent.
  • Google, user-triggered-agents.json, the live range file. The copy read on 18 October 2026 carried creationTime 2026-10-16T14:46:09.000000.
  • Google Search Central, technical requirements, including the condition that Googlebot is not blocked.
  • Google Search Central, AI features and your website, including supporting-link eligibility and the statement that indexing and serving are not guaranteed.
  • Google Search Central, optimising for generative AI features, including the Search Console include requirement, the separate note on browser agents, and the scaled content abuse warning.
  • OpenAI, overview of crawlers, including OAI-SearchBot and ChatGPT-User.
  • Perplexity, crawlers, including PerplexityBot and Perplexity-User.