As ordinary people get used to using AI chatbots, the services are increasingly evolving from a basic answer engine to more complex systems in their everyday lives.
According to the Reuters Institute for the Study of Journalism at Oxford University, daily use of generative AI systems across six countries surveyed nearly doubled in a year, growing to 28 percent in 2026 from 16 percent in 2025, with weekly usage jumping to 51 percent from 34 percent just a year ago. When respondents surveyed by YouGov were asked in August and September about generative AI usage from the prior week, the top three answers were “Researching a topic” (24 percent), “Asking advice” (20 percent) and “Answering factual questions” (18 percent).
“For the people who use conversational AI systems, they really do seem to be friction-reducing information machines and, perhaps increasingly, digital exoskeletons that act for them in the world — as we can see from the nascent use of AI agents,” said Felix Simon, a postdoctoral research fellow at the Oxford Internet Institute and one of the Reuters Institute report co-authors.
OpenAI’s own data shows ChatGPT use widening across both work and nonwork tasks. Last week, the company released data from a sample of 1 million monthly messages from individual ChatGPT accounts that showed writing accounted for only 21 percent of messages. Meanwhile, the largest category is now practical guidance comprising how-to advice, tutoring, brainstorming and health questions.
If you want to start leveling up your chatbot usage, here are some top things to try:
Dive into big topics
Features such as Deep Research — the name of tools from both OpenAI and Google Gemini — have been around for a while, but they’re a helpful way to use AI’s scale and speed to cast a wide net on a specific topic. Unlike a regular chatbot answer, both of these tools use a longer, multistep process. After breaking a user’s goal into smaller questions, the bots draft a research plan; search sometimes hundreds of sites, analyzing and adjusting along the way; and then finally write a report with citations. (The Washington Post has a content partnership with OpenAI.)
Another way to improve AI-powered research is connecting chatbots with outside datasets that enable chatting with court records on CourtListener, peer-reviewed research papers on Consensus, and financial data providers like S&P Global and FactSet. What’s accessible can vary and sometimes require subscriptions to access, but they help unlock entirely new sets or sources beyond just websites and training data.
To use OpenAI’s Deep Research, select the feature before you send a prompt. Write out the questions you’re looking to answer, the goal or project you’re working on, time periods and source types to consider, and the format for the report. After that, the chatbot probably will devise a plan for research and approval before running it.
It’s still important to verify sources. Also know that each chatbot has preferences for the websites it sources from and skips, often from a range of factors that include licensing deals, ongoing lawsuits, AI system instructions and technical details of specific pages.
Build a ‘second brain’
Instead of relying just on outside sources, connect a chatbot to your own notes. This might look like linking Claude Code to Obsidian or Notion note-taking apps, inboxes and Slack channels, browser extensions for clipping websites and saving bookmarks, or even locally stored files and folders on a laptop. This can become a more interactive way of using chatbots to find, write and organize documents.
For those wanting an easy place to start, Google’s Gemini Notebook lets people add up to 50 sources — such as documents, websites, voice files and even YouTube links — and then ask questions to analyze trends, spot outliers, create data visualizations, study guides or even create a brief podcast based on what’s in the notebook. While these have citations, it’s still key to verify your sources.
Stop re-explaining yourself
Most chatbots let you save context so you don’t start from scratch every single chat. For example, within Claude and ChatGPT, creating a Project becomes a designated home with its own custom instructions about your goal, your working style, when it should ask follow-up questions, how to format answers and specific ways you want AI to help. Projects also let you upload files that include more knowledge that the bot can pull from to help inform its answers and other reference materials. Other features include “skills,” sets of reusable instructions that a chatbot can load automatically when relevant.
Not sure what to write? Try “meta-prompting.” A prompt is the term for what you tell a chatbot to do, but meta-prompting uses a bot to write instructions for itself or for another by adding layers of detail that help it better understand a task. Try having the chatbot ask you questions before it drafts detailed instructions. Meta-prompting also helps improve content generated by image, video and audio tools by asking the chatbot for specific film or photography language, such as camera techniques, lenses, scenes, scripts and other aesthetic descriptions.
Make what you wish existed
Ask a chatbot to create something you wish existed, such as a video game, app, website or another tool. This is known as “vibe coding” and involves using plain language to describe an idea you have in mind and what you want it to do, which the chatbot then translates into code for building it. Some creations can run inside of a chatbot itself, but others can be published elsewhere for personal or public use on platforms such as Vercel. For example, I recently used Claude to make my own version of “Lemmings,” turning the classic 1990s video game into a web-based version with modern themes such as social media addiction, online misinformation and AI hype.
Chatbots also can help with visuals such as logos, websites, social media content and presentations. Beyond image- and video-generation tools within Gemini, platforms such as ChatGPT work with design tools such as Canva and Adobe Express.
For starters, create a meal-planning project with information about preferences, budgets, how many you’re feeding and other context. Or turn voice or chat AI into a personalized guide to help guide you through a museum, provide more information than the accompanying plaque or change voices to be more animated than a stuffy audio guide. Or create a Fantasy Premier League project and upload files with the official 2026-2027 rules, a tactics guide, performance history, fixture difficulty ratings and what you want help with — but have it act as a coach and scout rather than picking for you.
Other ideas
- Troubleshoot and fix things: One way more people are using chatbots is to help troubleshoot software, understand website errors, handle household repairs, or install new devices and appliances. An example could be uploading photos or screenshots to show what’s wrong with a certain issue online or offline. You can also use voice chats to explain in your own words what you’re trying to solve, or have a chatbot verbally walk you through something. However, it’s worth keeping in mind that voice dictation isn’t perfect yet and often misinterprets what a user has said. (Also make sure to check the model number for products to make sure you’re using the right manual.)
- Put something on autopilot: Heavy users are going far beyond answers. In the Reuters Institute survey, 14 percent reported using an AI agent — such as OpenClaw, Claude Code or OpenAI’s Codex — in the past week for real tasks that involve multiple steps across apps, websites, files, code, emails and calendars. Another 13 percent had only experimented, and 67 percent hadn’t used one.
- Decoding fine print: Chatbots also can help understand or explain complex jargon in technical and legal documents by turning it into plain language or extracting text to compare between versions of similar documents.
- Analyzing data: Chatbots can help extract text and data in PDFs, analyze spreadsheets and other types of data. They also can help plan trips by giving an AI agent arrival and departure times, interests and must-see spots, then building a realistic itinerary (or several).
The post Already using ChatGPT or Claude? These simple power-ups let you do a lot more. appeared first on Washington Post.
