A fifth of Google AI Overview traffic lands in Direct, nine-month GA4 study finds

One in five AI Overview referrals shows up as Direct

Google's AI Overviews can send meaningful referral traffic, but a nine-month first-party study suggests a substantial slice of it never appears where marketers look for it. According to an analysis published in Search Engine Land, based on a dataset from Alex Galinos, 22.4% of tracked AI Overview referral traffic was classified as Direct in Google Analytics 4 rather than Organic.

The study followed one transportation-industry brand from September 2025 through June 2026 and captured 51,200 AI Overview referral events across 1,661 cited snippets. The findings describe a single brand and should not be treated as a universal benchmark, but they illustrate a growing measurement problem: AI-mediated browsing does not always preserve a clean, end-to-end attribution path from click to report.

A volatile share of organic sessions

Across the full window, AI Overviews accounted for 7.53% of organic sessions for the studied site. That aggregate hides sharp swings. At its peak, AI Overview traffic represented roughly 16% to 17% of organic sessions, while more recent periods covered by the analysis show it receding to roughly 2% to 4%.

That volatility matters: a single month of reporting can make AI Overview referrals look like either a durable acquisition channel or a negligible one, depending on when the report is pulled. The study's implication is that teams evaluating whether content work is earning visibility in Google's AI results should avoid drawing broad conclusions from short reporting periods.

Monthly misattribution to Direct also varied, ranging from 16.8% in April 2026 to 29.3% in May 2026.

The text-fragment problem

A contributing cause, according to the analysis, is the text fragment (the format beginning with a colon, tilde and the word "text") that Google sometimes appends to destination URLs from AI Overviews. The fragment can take a visitor directly to the relevant passage on a page, but GA4 may then classify the resulting visit as Direct, as though the user had typed the URL, used a bookmark, or arrived through another unattributed route.

The pattern is not a definitive identifier, however. The same fragment format also appears on links from other Google features, including Featured Snippets, so not every Direct visit carrying it is an AI Overview click. The defensible conclusion is narrower: a portion of AI-influenced visits may be invisible in the channel where teams expect to find them.

GA4's AI Assistant channel is not a full fix

Google rolled out an AI Assistant channel in GA4 in May 2026, which improves how AI-driven visits are categorized. The analysis notes that it does not fully close cross-channel attribution gaps, particularly where referral information is not retained in a form analytics platforms can cleanly classify.

Measuring AI visibility as a set of signals

The study argues that assigning every sale or lead to a single AI Overview click creates false precision, since AI results can influence discovery, reinforce a brand choice, or send a visit that later surfaces under a different channel. It recommends separating AI traffic from AI influence and tracking a ladder of observable signals: discoverability (whether relevant pages are accessible, current and able to answer specific questions), search visibility (whether those pages rank for the queries customers use), and citations and mentions (whether AI Overviews cite or surface the brand's content).

Server logs can add a separate layer of evidence by showing bot and crawl patterns that browser analytics cannot capture, though they do not establish a causal chain from crawling to revenue. CRM self-reported attribution can help triangulate how a prospect first heard of a business. URL parameters remain useful for links a company controls, such as email, ads, social posts and partner placements, but they cannot be added to links Google independently places inside AI Overviews, which is why campaign tagging should be treated as one part of a broader measurement process rather than a fix for AI referral attribution.

On the content side, the dataset found that AI Overviews most often cited highly structured, specific pages answering concrete questions about times, prices, routes and named comparisons, and that content freshness mattered over time.

Why it matters

For anyone measuring organic performance, the finding means a standard GA4 channel report is an unreliable guide to AI-driven discovery. Roughly a fifth of AI Overview referrals in this dataset were filed under Direct, a channel that already functions as a catch-all, so teams relying on dashboards will systematically underestimate both their AI visibility and the contribution of content cited in AI-generated answers. That has direct budget consequences: if cited content appears to produce little traffic, investment in the structured, answerable pages most likely to earn citations becomes harder to justify. The practical takeaway is to pair analytics with citation monitoring, landing-page analysis, server-log evidence and CRM feedback, rather than expecting one channel report to carry the whole story.