Generative AI is getting to people via the software they already use, in what the Reuters Institute refers to as ”incidental” AI exposure. In a survey across nine countries, 57% of respondents had seen an AI-generated answer or summary in a search engine in the past week.
In the six countries tracked since 2024, 51% now use a standalone AI tool on a weekly basis (up from 34% a year previously). The battleground is dictated by distribution: who can get a useful model into search, into productivity software and into the services we use every day (Reuters Institute, 2026).
ChatGPT has the biggest audience of any of the dedicated tools at 34% weekly usage across the six countries with historical data. Gemini is second at 24% (up 13pp in the year) compared to ChatGPT's 12pp. Claude has gone from 1% to 7%, and Microsoft Copilot is at 11%. Google can introduce Gemini-powered features through search; Microsoft can embed Copilot into software people use at work. These routes reduce the effort required to start using AI and allow companies to reach people who may never choose a separate chatbot. In the nine-country survey, 25% encountered AI on phones or computers, and 21% through social or messaging platforms.
Weekly use of AI to get information rose from 11% in 2024 to 38% in 2026 across the six countries we have consistently surveyed. People are using it to do research, to get advice and to clarify facts. The report describes these as "friction-reducing machines," with 64% of those using it for information citing the benefit of getting faster, easier answers and 56% citing the conversational interface. The authors note that "trust and trustworthiness don't appear to be central" at this stage. Only 21% said they used it because they thought it was more reliable.
The report suggests users often settle for answers that are “good enough” for their purposes. As search and chatbot products compete to synthesize information and answer follow-up questions, retrieval quality, source citations and clear uncertainty become important design challenges. Despite all the attention around autonomous AI, only 14% of those surveyed had used an AI agent to carry out a real task in the previous week. Another 13% had tried one out, with the vast majority (67%) having not used an AI agent at all. The market as a whole is still very much in the 'get me an answer' and 'produce me content' phase rather than actually delegating workflows to software that can act on your behalf.
Access to the more advanced capabilities is also very uneven. 10% paid for AI and 9% had access via work or education (allowing for overlap, this is 18%). Most people still use the free versions, though, a useful reminder that the latest model/agent demo you've seen is not necessarily what most people actually use.
Tech companies ought to make AI useful where work and information-seeking already take place. Search, browsers, operating systems, and workplace platforms offer that reach. Autonomous agents will need to prove they can complete tasks reliably before they become comparably routine.