Traffic estimates show Bing falling while Google and ChatGPT grew

Semrush estimates for July 2026 show Google and ChatGPT gaining US website visits while Bing lost roughly half of its traffic year on year. The comparison is worth examining, but it does not establish that ChatGPT or Copilot took those visits from Bing.

What the numbers are, and what they are not

The figures come from Search Engine Land’s rundown of US web traffic, built on Semrush’s Traffic Analytics tool, comparing July 2026 against July 2025.

One caveat up front, because it changes how much weight to put on any single line. Nobody outside these companies can see their real server logs.

Tools like Semrush and Similarweb estimate traffic from panels — samples of real browsing behaviour — and then scale up. Sampling and modelling introduce uncertainty, so avoid treating the decimal places as exact counts. Treat “Bing fell about half” as the finding, not “Bing fell 50.43%”.

Here is what the estimate says for the US, July on July:

• Google: 25.3 billion monthly visits, up 10.72%
• YouTube: 10.3 billion, up 36.6%
• ChatGPT.com: 1.09 billion, up 48.38% — now ninth overall
• Bing: down 50.43%, the worst decline anywhere in the top 20
• DuckDuckGo: down 15.83%

The figures do not explain the cause

One possible factor is the shift between website visits and activity inside apps.

Some searches may now happen inside Copilot rather than on bing.com. Microsoft has been pushing Copilot into Windows, Office and Edge, so the query gets asked inside an app rather than by typing into bing.com. That query never registers as a website visit at all. The traffic estimates do not quantify how much of Bing’s decline this explains.

Google can also deliver AI answers within its existing search service. Website traffic therefore cannot cleanly separate conventional search from AI-assisted search.

Generative-AI traffic is spread across several services

A separate dataset describes changes within the generative-AI category.

Similarweb’s generative AI traffic figures cover June 2025 to May 2026, worldwide. The whole category grew about 70% year on year, to an average of 9.5 billion monthly visits. Inside that category, the shares moved hard:

• ChatGPT went from around 76% of gen-AI web traffic to about 53%
• Gemini went from under 9% to roughly 27–28%
• Claude went from about 2% to close to 9%

ChatGPT is still the biggest by a distance. That is a reason to name the assistants included in a report. Visits to generative-AI sites are not all search activity, and this worldwide dataset should not be combined directly with the US estimates above.

What belongs in a client report

Keep technical SEO in the review, but check each provider’s access separately. Google’s AI-search guidance retains its search eligibility requirements. OpenAI documents its own search crawler, OAI-SearchBot, with separate access controls. A Google crawl failure alone does not establish whether ChatGPT can retrieve a page.

Report each measured assistant separately. A brand can appear frequently in Gemini answers and be absent from the Perplexity sample. An aggregate hides that difference unless the report also shows the services and prompts included.

Keep AI referral traffic alongside direct citation measurements. Google Analytics added a dedicated AI Assistant channel in May 2026, but that channel cannot identify every visit originating in an AI answer.

But as Search Engine Journal noted at the time, traffic that arrives without a referrer header — the small note a browser passes along saying where the click came from — still lands in Direct. Some app and in-app-browser journeys omit it. Copy-pasted links never had it.

Do not treat the recorded referral count as the total AI audience or estimate the missing visits with a fixed multiplier. Search Console’s AI reporting measures appearances in answers separately from clicks.

The workable answer is to measure presence in the answers themselves: ask each engine the questions your customers ask, and record whether you come back.