AI Search can be measured in several ways, and the latest Semrush data shows that the results can look completely different depending on which metric you examine. In the sector studied, only two domains appeared in both the top 15 most frequently mentioned brands and the top 15 most frequently cited sources. And all traffic sent directly by AI tools accounted for just 0.48% of sessions.

First, an important limitation: this is not a result for the entire internet

Semrush studied the US manufacturing and industrial sector, not every industry. The dataset covered 458 terms related to industry and manufacturing, with data collected from January to July 2026. Traffic was analysed using US sessions across ten industrial categories. AI visibility was compared across ChatGPT, Gemini, Google AI Mode and AI Overviews.

There is therefore no basis for applying the 0.48% figure to a clothing store, a law firm or a local service business. There is, however, a very good reason to treat the study as a demonstration of a measurement problem that affects practically every company trying to assess its presence in AI today.

WebsiteGarage / Services

Explore our services.

AI Overviews are expanding, but traffic from AI is still small

Within the keyword set studied, the share of search volume for which Google showed an AI Overview rose from 38% in January to 57% in July. That does not mean AI Overview appeared for 57% of all Google queries. Semrush weighted the result by search volume within its set of 458 industrial terms.

At the same time, AI Mode and AI assistants together accounted for 0.48% of sessions sent to the websites studied. Traditional organic search and direct visits still made up the overwhelming majority of traffic.

This is a good example of why the statement “AI is taking over search” can be both true and completely useless. AI can occupy more and more space in the search process without sending proportionally more clicks to source websites.

A mention, a citation and a click are three different things

Semrush defines a mention as a situation where a brand name appears in an AI answer. A citation means the system points to a specific website as a source and may provide a link to it.

In the study, the leaderboards for those two categories barely overlapped. Of the fifteen most frequently mentioned brands and the fifteen most frequently cited domains, only Vevor and Grainger appeared in both groups.

That matters because a brand can be highly visible in AI answers thanks to reputation, recognition and information from many other sources, even if its own website is not cited very often. Conversely, a website can be an excellent source for a model and receive many citations without building an equally strong brand presence.

ChatGPT and Google do not use the same source ecosystem

One of the more interesting observations concerns engineerfix.com. In Semrush’s data, it was cited exceptionally often by ChatGPT, but did not appear among the most frequently cited sources for Google Gemini, AI Mode or AI Overviews.

This does not prove a simple causal relationship, but it shows something very practical: “AI optimisation” is not a single channel. Systems use different search mechanisms, partnerships, indexes and sources. A strong result in one tool does not have to repeat in another.

The most interesting question is: why should the user click?

The study also shows the difference between content that is easy to summarise and content that cannot sensibly be replaced by a few sentences. Semrush compares general reference material with original scientific publications from the Royal Society of Chemistry. The latter gave users a clearer reason to visit the source.

That may matter more than the next trick involving headlines written “for AI”. If a page only answers a question that a model can fully summarise in four bullet points, a citation does not have to lead to a click. Original data, a tool, calculator, offer, configurator, documentation, comparison or material based on first-hand experience creates something an AI answer cannot simply reproduce.

WebsiteGarage’s take

The most useful conclusion from this study is not “AI brings only 0.48% of traffic”. That number applies to a specific sector and a specific sample. More important is the gap between three stages: AI can know of a brand, cite its website and send a user onwards. Those are not synonyms.

That is why a report saying “ChatGPT mentioned us 137 times” without context is about as useful as the number of ad impressions without any information about whether anyone did anything afterwards. In AI Search, brand presence, citation sources, traffic, branded queries and conversions should be measured separately. Only then can you see whether visibility is an interesting statistic or actually part of the customer journey.

And there is one more piece of good news. We do not need to reinvent the internet just because language models appeared. Content with its own value, that is specific, credible and gives users a reason to visit the source, still makes more sense than producing ten generic articles a day in the hope that one of them appeals to a bot.