Semrush’s July 2026 snapshot of U.S. web traffic shows a deceptively stable top of the market—and a fast, uneven redistribution underneath it. Google and YouTube still dominate total visits, but AI-native properties such as ChatGPT have moved into the top 10, some legacy search engines have lost large shares, and GA4’s default attribution is increasingly unable to explain where audiences actually come from.

What moved: concentration rose, AI chatbots climbed

At a glance the rankings look familiar: Google, YouTube and Reddit remain the three most visited U.S. sites. Look closer and the shifts matter. Google’s visits rose 10.7% year over year to 25.3 billion monthly visits; YouTube jumped 36.6% to 10.3 billion. The top 10 sites now produce roughly 68.6% of visits across the top 150, underlining growing concentration at the highest ranks.

Within that pattern, ChatGPT.com was one of the clearest winners. It grew about 48.4% year over year to roughly 1.09 billion monthly visits and now ranks ninth among U.S. domains—outranking some long-established search engines. That rise signals that generative AI moved from novelty to habitual use fast enough to register as one of the largest traffic gains in the top 20.

Other notable gains: LinkedIn entered the top 20 (up about 15.5%), Twitch climbed after a 25.6% increase, USPS.com rose modestly, and smaller sites such as Comix.to posted outsized percentage gains. Those winners underscore that utility and entertainment continue to attract audiences even as discovery patterns shift.

Who lost ground, and why it matters

Some declines were striking. Bing fell about 50.4% year over year—despite Microsoft’s heavy investment in Copilot—suggesting AI feature integration has not translated into standalone traffic growth. Temu’s traffic dropped 28.6% after the end of the U.S. de minimis tariff exemption in mid-2025 removed a core pricing advantage, an explicitly policy-driven decline rather than an SEO failure.

Other losers include Amazon (down ~15.3%), DuckDuckGo (down ~15.8%), NBCNews.com (down ~20.2%) and ESPN (down ~16%). These declines are not uniform in cause, but together they show traffic reallocating unevenly below the top few winners. The simple headline “Google and YouTube still win” is true, but it masks important redistribution driven by AI usage, policy changes and shifting consumer habits.

Why GA4 attribution is fraying

The timing of these shifts matters because Universal Analytics stopped processing new data on July 1, 2023. Many teams now rely on GA4 defaults for channel performance—but those defaults are a weakening lens for AI-driven referrals.

An anonymized case illustrates the issue: one niche publisher’s GA4 view comparing Jan 1–Aug 26, 2026 with the same period in 2025 showed sessions and engaged sessions up more than 100%. Yet GA4 labeled over 78% of traffic as “direct,” up from about 62.5% the prior year. “Direct” in GA4 functions as a catch-all for unattributed visits; an increasing share of that category likely represents traffic arriving via AI chat interfaces or other referrer-stripping environments that remove the metadata GA4 depends on.

GA4 added an “AI Assistant” channel in May 2026, but native reporting attributes only a fraction of expected AI-driven traffic there. That gap aligns with independent analysis earlier in 2026 arguing GA4 alone cannot reliably measure AI-impacted acquisition paths.

What marketers should do now

Two practical shifts are immediately useful:

  • Treat GA4’s default channel mix skeptically. Large “direct” shares increasingly mask origin; use them as a signal of missing attribution rather than evidence of direct loyalty.
  • Widen measurement beyond GA4. Add audience intelligence and panel tools that estimate visits and referral composition—examples used by traffic analysts include Semrush’s traffic estimates, SparkToro for audience affinities, GWI for panel-based segmentation and YouGov Profiles for continuous survey data.

Complement external tools with first-party voice-of-customer research. Short surveys (SurveyMonkey, Typeform, Google Forms, Qualtrics) can surface what channels audiences are actually using when analytics go dark. When analytics can’t show where an audience came from, ask the audience directly and use those answers to prioritize channel tests.

What to watch next

Watch whether AI-native properties continue rising, whether Microsoft can arrest Bing’s decline, and whether Google’s AI product changes further strip referrer signals. For practitioners, the immediate priority is a hybrid measurement posture: use GA4 for behavioral and engagement metrics, but build decisions on a broader evidence base that includes audience intelligence, panels and direct surveys. That approach trades perfect attribution for a clearer, testable picture of intent and distribution—essential while referral metadata grows less reliable.