Daily AI chatbot use nearly doubled in a year, Oxford study finds

The latest:

Daily use of generative AI systems nearly doubled in a year across six surveyed countries, rising to 28 percent in 2026 from 16 percent in 2025, according to the Reuters Institute for the Study of Journalism at Oxford University. Weekly use climbed to 51 percent from 34 percent. The findings land as OpenAI’s own usage data shows practical guidance overtaking writing as the largest category of ChatGPT messages.

Details:

• The survey: The Reuters Institute figures are drawn from YouGov polling conducted in August and September across six countries. Respondents were asked about use in the previous week, a shorter recall window than annual-use questions, which tends to capture habitual rather than occasional users.
• Top uses: Researching a topic led at 24 percent, followed by asking advice at 20 percent and answering factual questions at 18 percent. All three leading answers involve information retrieval or decision support rather than content creation, the category chatbots were first marketed on.
• The researcher’s reading: Felix Simon, a postdoctoral research fellow at the Oxford Internet Institute and a co-author of the report, described conversational AI systems as friction-reducing information machines and, increasingly, digital exoskeletons that act for users in the world, pointing to early uptake of AI agents.
• OpenAI’s data: OpenAI last week released findings from a sample of 1 million monthly messages from individual ChatGPT accounts. Writing accounted for 21 percent, while the largest category was practical guidance, bundling how-to advice, tutoring, brainstorming and health questions. The company did not break out that category’s share.
• Deep research tools: OpenAI and Google Gemini both offer features named Deep Research, which break a goal into smaller questions, draft a research plan, search sometimes hundreds of sites while adjusting along the way, then produce a report with citations. The feature must be selected before a prompt is sent.
• External datasets: Chatbots can be connected to outside data sources including court records on CourtListener, peer-reviewed papers on Consensus, and financial providers such as S&P Global and FactSet. Access varies and some connections require paid subscriptions, extending the systems beyond websites and training data.
• Source limits: Each chatbot favors and skips particular websites for reasons including licensing deals, pending lawsuits, system instructions and technical characteristics of individual pages. Verifying cited sources remains necessary even when a tool returns citations. The Washington Post has a content partnership with OpenAI.
• Personal archives: Users can link chatbots to their own material rather than only external sources, connecting Claude Code to note-taking apps such as Obsidian or Notion, to inboxes, Slack channels, browser clipping extensions or local files, turning the tools into interactive systems for finding, writing and organizing documents.
• Notebook option: Google’s Gemini Notebook accepts up to 50 sources, including documents, websites, voice files and YouTube links, then answers questions on them, identifies trends and outliers, and generates data visualizations, study guides or a short podcast drawn from the uploaded material.
• Saved context: Claude and ChatGPT both support Projects, a persistent space holding custom instructions on a user’s goal, working style, preferred answer formatting and when the bot should ask follow-up questions, along with uploaded reference files. Skills are reusable instruction sets loaded automatically when relevant.

Between the lines:

The usage pattern and the product features point the same direction. With research, advice and factual questions leading weekly use, and writing down to 21 percent of ChatGPT messages in OpenAI’s sample, the heavier investment in multistep research tools, external dataset connections and persistent project memory tracks what people are already doing rather than steering them somewhere new.

What’s next

Watch whether the Reuters Institute’s next annual survey shows daily use continuing to double, and whether OpenAI publishes a breakdown of the practical guidance category it now identifies as the largest share of ChatGPT messages.