AI Overview Traffic May Be Hiding in Direct: What 51,200 Referral Events Show

Google's AI Overviews are generating referral traffic, but current analytics reports may not capture the full extent of this traffic. A comprehensive analysis of 51,200 AI Overview referral events from September 2025 to June 2026 revealed that 22.4% of this traffic was incorrectly classified as Direct traffic in Google Analytics 4 (GA4) rather than being attributed to Organic search.

This discrepancy is significant because teams that rely solely on channel dashboards may underestimate both the visibility of their content and the impact of AI Overview citations.

The nine-month dataset, published by Alex Galinos in Search Engine Land's analysis, focused on one transportation-industry brand and included 51,200 AI Overview referral events across 1,661 cited snippets. The findings highlight the volatility and unpredictability of AI Overview traffic, which can fluctuate dramatically from month to month.

At its peak, AI Overview traffic accounted for 16% to 17% of the site's organic sessions, while in other periods, it dropped to a mere 2% to 4%. This variability means that short-term reporting periods can lead to misleading conclusions about the lasting contribution of AI Overview traffic.

One of the key issues identified is the way Google sometimes appends a "#" followed by a fragment (#:~:text=) to the destination URLs from AI Overviews. This fragment can send users directly to specific text on a page, but GA4 may classify these visits as Direct traffic because the user might have typed the URL, used a bookmark, or arrived via another unattributed route.

While this is not a definitive indicator of AI Overview traffic, it suggests that a portion of AI-influenced visits may be invisible within the channel where marketers typically expect to find them.

Google's May 2026 update introduced an AI Assistant channel in GA4, which improves the categorization of AI-driven visits. However, this does not fully resolve the cross-channel attribution gaps, especially when referral information is not retained in a way that analytics platforms can easily classify. To accurately measure AI visibility, businesses should adopt a more nuanced approach that goes beyond assigning every sale or lead to a single AI Overview click. Instead, they should focus on tracking a series of observable signals:

1. **Discoverability**: Assess whether relevant pages are accessible, up-to-date, and capable of answering specific questions.

2. **Search Visibility**: Evaluate whether these pages rank for the queries customers use.

3. **Citations and Mentions**: Determine if Google AI Overviews are citing or surfacing the brand and its content.

4. **Server Logs**: Utilize server logs to capture bot and crawl patterns that browser analytics cannot, providing additional evidence of how search systems access content.

5. **CRM Attribution**: Use CRM self-reported attribution to triangulate how a prospect first learned about the business, particularly when analytics data is incomplete.

While URL parameters can be valuable for links businesses control, such as campaign links in email, ads, social posts, and partner placements, they cannot be added to links placed by Google in AI Overviews. This limitation underscores the importance of clean campaign tagging as a single component of a broader measurement process, rather than a solution for AI referral attribution.

The core question for businesses should not be whether every AI-generated visit can be perfectly labeled, but rather whether their content can be discovered, cited, and connected to commercial results through multiple independent signals. AI search visibility is becoming increasingly complex to assess from a single GA4 channel report.

Tools like GEO Search Leads can assist businesses in identifying where their content is visible in AI-driven discovery and highlight opportunities that may be missed through incomplete Direct or Organic data. By focusing on these multi-faceted signals, businesses can better prioritize their content strategies and avoid the pitfalls of incomplete attribution.