Blogs

How to Measure Google-Extended

Google-Extended is a robots.txt token for Gemini training, not a second Googlebot. Measure Googlebot crawls separately and do not expect a Google-Extended ads pixel.

Ecem Bircan
Data Analyst
Marketing

Google-Extended is the robots.txt product token Google uses so you can opt out of Gemini / generative-AI training without opting out of Search.

Google has been explicit that Search crawling stays on Googlebot, and AI Overviews and AI Mode still ride that search crawl. Blocking Google-Extended does not remove you from Search, and it does not create a "Gemini Ads" conversion API.

If your log dashboard has a line named Google-Extended, read it as "training-control token / related hits," not as a campaign. The crawl you already know how to measure is Googlebot.

What "measuring" it means

  1. Policy: Is User-agent: Google-Extended present in robots.txt, with an explicit Allow or Disallow? Ambiguous blocks get ignored.
  2. Search crawl: Googlebot volume and Coverage in Search Console, unchanged by the Extended token.
  3. Hardal: AI Visibility lists Google-Extended among named agents so you do not dump it into a generic bot bucket. Use it to confirm the training opt-out is the one you intended.

Do not look for Google-Extended in GA4 acquisition. There is no pageview to find.

robots.txt

Opt out of Gemini training, keep Search:

User-agent: Google-Extended
Disallow: /

Leave Googlebot allowed. Disallowing Googlebot does nothing about Gemini. It removes you from Search.

This is the rare clean split in the AI bot list. GPTBot vs OAI-SearchBot is a similar idea at OpenAI. Google just happens to reuse Googlebot for search-shaped AI features, so the training switch is a token rather than a second spider you will see as often as GPTBot.

What not to build

  • A "Google-Extended Conversions API" guide. There isn't one.
  • A ChatGPT Ads clone with the name swapped. Paid Google measurement is still gclid, Enhanced Conversions, offline imports, and server-side GA4 / Ads, which Hardal already treats as destinations, not as this token.

For ChatGPT paid media, start at How to measure ChatGPT Ads.