How people are using GenAI chatbots: Evidence from web traffic data

Tracking the uptake of artificial intelligence (AI) poses a fundamental measurement challenge. Recent data, such as Eurostat’s survey, are useful but often limited in geographic and chronological coverage and can be difficult to adjust as technologies change. Given that AI usage is evolving rapidly, frequently updated data sources can help us understand the dynamics and address data gaps.

To measure AI usage in real time, the OECD.AI team has developed an approach that leverages web traffic data. Platforms such as Similarweb provide estimates of website and app visitors, offering a window into how users interact with digital services at scale.  This web traffic data enables researchers to track usage trends for AI chatbot interfaces and services.

While these data should be interpreted with appropriate methodological caveats, they offer unique insights into real-time dynamics of how the public is using AI.

To build a measure of monthly usage of GenAI chatbots, we focus on direct user engagement through AI web interfaces. It uses the number of unique, deduplicated visitors over a two-year period between 1 February 2024 and 30 March 2026 to the three main AI chatbot websites, ChatGPT, Claude and Gemini. According to the data, these three sites account for essentially all web traffic to chatbot interfaces in GPAI countries, as users of other sites will almost always also use one of the three leading sites and would therefore be removed during the deduplication process anyway.  

With the number of unique visitors across the three chatbots, divided by population from IMF projections, the new measure of GenAI chatbot usage is computed.