How to Identify all LLM Referrers

[Joseph_Making]

Hi 🙂 

I was looking to identify all LLMs referring people to our site. For many they have unique referring domains, and Chat GPT seems to add their own query string parameter to URLs (thanks Open AI lol), but for things like google search ai summary, does anyone have any recommendations?

Thanks! 

[bjoern__koth]

Hi @Joseph_Making 

Adobe just recently added a new referrer type "Conversational AI" that should roll out any time. See also: "Analyze AI traffic with a new Referrer Type dimension item".

Unfortunately no documentation yet, and from a quick check in a client's report suite, there I could not yet see any data.

[Jennifer_Dungan]

@bjoern__koth and @Joseph_Making, I believe the new "Conversational AI" will be fully released in the big Q4 Release on Oct 16th, 2025.

It will not be retroactive, since that would be detrimental to historical reporting.

From the Release Notes:

In October, a new Referrer type dimension item will be available to help analyze traffic that comes from AI tools.

This new Referrer type dimension item, called Conversational AI tools, will group together referring domains of major AI tools, allowing you to look at trends for the group as a whole. The initial list of referring domains in this new category includes (but is not limited to):

• chatgpt.com
• claude.ai
• m365.cloud.microsoft
• grok.com
• gemini.google.com
• perplexity.ai

The new dimension item will be available in all Adobe Analytics-related tools including Analysis Workspace, Report Builder, Data Warehouse, Data Feeds, and so forth.

Consider the following when using this new dimension item:

• It is not always possible to distinguish referrer traffic that came from results provided in “AI mode” in a search engine vs. those that came from click-throughs from traditional search results.
• The new Conversational AI tools dimension item focuses on major providers with the most traffic. A new trend reveals a rising number of impersonator sites with domains similar to major AI tool providers. This is likely due to the ease with which individuals or groups can create their own AI tools and host them on the internet. Because this is a rapidly evolving space, please contact the Adobe support team if you find that a popular site is not included.
• The Referrer type dimension, including the new Conversational AI tools dimension item, is available only for data that is processed by Adobe Analytics.

(Documentation link to follow.)

---

 If you need to get data on your current data set, you could create your own segment looking for known referrers, and the ChatGPT UTMs.

However, the Google Search AI Summary is less than ideal... there is no true identifier... the closest thing is the use of Google as a Referrer in combination with the use of a Scroll to Text Fragment...  

Scroll to Text is not unique to AI Summary links.. but the combination of it with the Google Referrer could get you close...

Scroll to Text Fragments are more protected than normal fragments, meaning you can't extract them with normal Fragment code, but they are available via the performance object.

Someone wrote a blog post to extra these types of fragments for use with GTM and GA4, but this can be modified for use with custom code Data Elements.

https://www.owntag.eu/blog/scroll-to-text-fragments-in-ga4/

But again, I would be mindful to pair this specifically with Google as a referrer, or maybe even specifically paired with Google Search: collecting scroll to text fragments in a dimension (or just a flag indicating the use of scroll to text), and creating a Marketing Rule (above the existing Search rules) to pair Natural Search Detection Rules with the scroll to text fragment?

[Jennifer_Dungan]

Oh, I should mention.... the Scroll to Text Fragments are used by AI Mode... not by the AI Summary... (I try to avoid using both of those)... I cannot see any way to properly determine traffic specifically from the AI Summary.

But the steps above should work for the AI Mode.

[manojsingh-tech]

Alternative Method Using Segments:

You can also approach this by creating a segment at the visit level using two containers, both set to the visit level. Each container should use an OR condition, and the overall logic between the two containers should also be OR.

• Container 1 (User Agent):
  Set the condition to:
  User Agent contains

  • openai
  • chatgpt
  • google-extended
  • anthropic
  • perplexity
  • claudebot
  • ccbot
  • bytespider

• Container 2 (Referrer):
  Set the condition to:
  Referrer contains

  • .ai
  • perplexity
  • gemini

This setup should help you capture traffic directed from known AI bots.

Please note: This method won't capture all LLM-related requests to your site. Many LLMs access content at the HTML level, which prevents Adobe Analytics from tracking them properly since JavaScript doesn’t get executed in those cases.

Hope this helps!