AI referral traffic is the part of AI visibility that shows up in a company’s own web analytics. For example, someone asks ChatGPT to recommend staff scheduling software, the answer names a provider and gives its product page as a source, and a click on that link becomes a visit to the page. “Referral” is the web analytics term for a visit that another website sent your way. AI referral traffic therefore requires a link in the answer, such as an AI citation. Fetches by AI crawlers, or by AI systems that read a page on a person’s behalf (user-triggered fetchers), are not part of it: no person visits the website.

Mentions and citations happen inside the AI answer. Visits from links in the answer, by contrast, can be analyzed like any other channel: you can see what visitors do, for example whether they send an inquiry or make a purchase. This makes the business value of AI visibility tangible.

How web analytics recognizes visits from AI answers

Web analytics tools assign every visit to a source. They rely mainly on two signals:

  • Referrer: When someone clicks a link, the browser usually tells the destination page which address the visit came from. This information is called the referrer. When a link leads to another website, browsers send only the domain of the linking site, such as chatgpt.com, unless the linking page specifies otherwise. The address of the individual page is not passed on. This is the default set out in the “Referrer Policy” specification of the W3C, the body that develops web standards.
  • UTM parameters: These are additions at the end of a link address that analytics tools read. Their names start with utm_: utm_source names the source, utm_medium the type of channel, and utm_campaign a campaign. According to OpenAI, ChatGPT automatically adds utm_source=chatgpt.com to links to sources, for example in answers that use ChatGPT search. A link to example.com/scheduling then becomes example.com/scheduling?utm_source=chatgpt.com.

Based on these signals, analytics tools group visits into channels. Since May 13, 2026, Google Analytics has shown a separate “AI Assistant” channel in its Default Channel Group, the built-in grouping of visits by source. Google assigns visits to it when the referrer matches a list of recognized AI assistants, naming ChatGPT, Gemini, and Claude as examples. The analytics tool Matomo also groups visits from AI assistants by referrer in its own AI Assistant channel. In other tools, such visits can be filtered by the referrer domains of AI systems, such as chatgpt.com, and by utm_source=chatgpt.com.

Why the numbers can be too low

Not every visit from an AI answer is recognized as one. If there is neither a referrer nor a UTM parameter, web analytics cannot assign the visit to a source and counts it as direct traffic, as if someone had typed in the address. This happens, for example, when someone copies a link without UTM parameters from an AI answer and pastes it into the browser, or when the linking page specifies that no referrer be sent. Under the Referrer Policy specification, the referrer is also dropped when a link leads from an encrypted page (https) to an unencrypted one (http). In addition, analytics tools only recognize the AI systems on their lists. Visits from other AI systems appear as ordinary referrals from a website.

Measured AI referral traffic is therefore a lower bound rather than the full figure. How large the gap is cannot be quantified in general.

What AI referral traffic says about AI visibility

AI referral traffic shows only part of a brand’s AI visibility. AI answers can name a brand without linking to it (a plain mention), and even a linked source is not clicked every time. An analysis by Pew Research Center of the browsing data of 900 US adults in March 2025 found that on Google pages with an AI-generated summary (an AI Overview), users clicked a link in that summary in just 1% of visits to such pages; only clicks on the first three sources cited in a summary were counted. The analysis covers only Google and the US, but it shows that an AI answer is often not followed by a visit. Whatever such an answer says about a brand then does not show up in any web analytics.

Conversely, Google’s and Microsoft’s reports on their AI features do not count visits: the generative AI performance report in Google Search Console shows how often links to a website were displayed, and according to Microsoft, a citation in the AI Performance report in Bing Webmaster Tools does not represent traffic or clicks.

AI referral traffic is therefore most meaningful alongside other metrics. In GEO monitoring, it sits next to mentions and the citation rate: the citation rate shows how often AI answers give a website as a source, and AI referral traffic shows how many people actually come as a result. Recorded monthly, together with the inquiries or purchases that follow, it shows the trend over time and is easy to report to management.