Semrush counts ChatGPT in visits and Bing in referrals in the same July table

A visit and a referral are not the same event. Semrush's chatgpt.com figure counts visits its panel observed on the site, the Bing figure counts referrals, and the Google figure is reported as direct ^[1][2][3]. Semrush's own caveat in the analysis says the table combines different signals reported within the same analysis, including visits and referrals, and should not be read as a like-for-like market-share comparison ^[7].

Take the growth figure at face value and it implies a baseline. Dividing 1.09 billion by 1.4838 puts last July near 735 million visits, so the year added roughly 355 million ^[1]. The input is a clickstream projection across desktop and mobile from a panel Semrush says includes roughly 200 million anonymized users ^[4][5]. So it is an estimate. Semrush states the figures are not first-party traffic reports from OpenAI, Google, or Microsoft ^[6].

Concentration is the part of the July data a publisher should care about more than the growth rate. Semrush put roughly 21.6 percent of ChatGPT referrals into Google and more than 30 percent into the top ten domains ^[9]. Divide those: Google alone is at most 72 percent of the top-ten block, and the nine other domains split at least 8.4 points of total referrals between them ^[2][3]. Semrush's April 2026 study reported that ChatGPT outbound referral traffic grew 206 percent during 2025 ^[8].

The attribution problem is in how the referral gets passed along, not in the chat model. Referral data depends on how the originating surface passes traffic to a browser and how that visit is captured, and in AI chat interfaces those pathways may not always appear as a clean, consistent referral source ^[11]. The analysis lists the arrival paths: a link, a copied URL, a follow-up search, or another path that obscures the original interaction ^[12]. Only the first of those has much chance of carrying the chat as its origin. Semrush did not quantify the gap; its advice is that a low or missing AI referral count is not proof that AI discovery is irrelevant ^[12][4].

For the 48.38 percent to mean anything inside your own dashboard, two things have to hold. The panel has to resemble your audience, and your channel definitions have to match the ones Semrush mixed in that table. You cannot check either one from outside the vendor.

GA4 still measures owned web activity, and the analysis is explicit that AI-driven referrals create attribution challenges on top of it ^[14]. The suggested repair is to read GA4 acquisition reports next to panel estimates treated as directional, and to track branded search interest, referral patterns, and landing-page performance together instead of leaning on one channel label ^[13].

Semrush's referral findings also suggest ChatGPT sends meaningful traffic toward Google, so a session can start with a conversational question and finish on a search result ^[15]. Google remained the largest referrer in Semrush's analysis, and its reported growth was positive ^[16].