Three years after ChatGPT’s release, two major camps have formed: those who refuse to use ChatGPT and those who use it every day.
In a 2025 survey by Pew Research Center, one-third of U.S. adults said they use ChatGPT. This includes 58% of U.S. adults under 30, nearly double the share from two years ago.
Experts say open conversations about how to best use AI are becoming increasingly important as a gulf is emerging between those already deeply invested in AI and those who refuse to engage with it.
So how should you approach AI tools? Here’s a step-by-step guide from experts.
brainstorming ideas
Deciding when to use AI can be difficult if you are not yet familiar with its capabilities. You can start by simply asking for help with something that’s already on your mind or on your to-do list.
Timothy B. Lee, author of Understanding AI Newsletter; We recommend using AI tools to brainstorm and break down tasks and projects into achievable steps. “Anytime you’re trying to come up with an idea, that’s a good starting point,” he says.
Catherine Goetze, a content creator and AI educator who goes by @askcatgpt on TikTok, suggests thinking of TikTok as a “thinking partner.” This can help you bounce ideas off, break through creative blocks, and refine your thinking. However, when reviewing results, it’s important to continue to use your own judgment, expertise, and preferences rather than letting the AI have the final say. The best tasks for AI are those where “we know what the right answer is,” she says.
research project
For more complex or intensive research, AI tools can be used to obtain an overview of what has been published. “Think of it like Wikipedia,” Lee says. “We know that’s false and we know how to check citations.”
Tools such as Claude, ChatGPT, and Perplexity all offer variations on “Deep Investigation” functionality. This will scrutinize the documentation on your topic and summarize it into a few thousand word report. Bots may also ask clarifying questions to collect the most relevant results.
“It’s really amazing,” Lee says. The response will include primary sources and links for you to refer to for yourself. Lee says the deep research feature helps him “get the picture” that identifies important papers to read, questions to answer, and next steps for a project.
Experts agree that’s the key point. Even after using AI tools for preliminary orientation, you still have to do the actual work yourself.
learn a new skill or hobby
Associate Professor Ella Havelmarz of the Vrije Universiteit Amsterdam, who studies the impact of generative AI on work, says AI can also be used “when it can augment your world.”
Havelmalz not only uses it to grow tiny lotus flowers, but also learns the basics of stock investing and even generates ideas for what to make for dinner.
The benefit, she says, is that it “kick-starts” new interests, activities, hobbies and skills, especially for those with high barriers to entry.
“For many people, the hurdles when trying something new are embarrassment, fear, time, discomfort, and not even knowing what to know,” she says.
Interacting with AI can break down these barriers and “allow us to take the next step,” says Havelmalz. But again, this is probably best treated as a starting point for tasks that are low risk and allow you to maintain ultimate authority.
Organizing information
Once you start a research project, AI tools can help you structure your research results by identifying themes, answering questions, generating timelines and summaries, and more.
If research is a priority, Hafermalz recommends using Google’s free application NotebookLM. NotebookLM only retrieves information from the documents, notes, and materials you upload, not from the entire web.
According to Hafermalz, historians are currently using NotebookLM as a research assistant. She continued: “It works as an organizer…If you’re trying to organize information and consolidate things and you don’t want it to go off the rails and pull things out of Reddit, NotebookLM is a more restrained space.”
For personal use, AI acts like a housekeeper or executive assistant, helping you plan meals and training, create a budget, and structure your afternoon from high-priority to low-priority tasks.
How to get better results
While creating prompts was important a year or so ago to get the most out of an AI tool, “over time, it’s become less and less important,” Lee says.
However, there are still some things you can do to get the most out of them.
You can now trust cutting-edge AI tools to respond more intuitively and to more casual words and expressions. However, context also helps. “The more information you provide, the more likely you are to get a good outcome,” Lee adds.
Götze encourages people to abandon the idea of prompts altogether. “Think of it as chatter,” Goetze says. “The magic really comes from the back and forth.” After all, that’s ChatGPT’s advantage over Google Search, she adds.
You may also link to your website and upload PDFs or other materials to reference in your answers, as long as you are careful to avoid sharing sensitive information. For example, you can ask AI to share your new phone contract and flag any terms you should be aware of or potential savings opportunities.
If you get stuck on something, you can also do something called a “reverse prompt.” Götze recently asked ChatGPT to come up with five questions to help her overcome mental blocks when she’s working on a document. “They inspired me to think a little differently,” she says.
It’s important to review all AI responses. Experts agree that while things are getting better every day, they still need to be checked against reliable primary sources. In addition to spouting falsehoods and misinformation, this model can also actively make things up, a phenomenon known as “hallucinations.”
Goetze says, “Check the source, check the links, check the date of the source.”
AI can also be used in ways that shrink our world and undermine our capabilities and humanity. They often result from treating them as shortcuts, or the final stage of a process, rather than the first step.
Götze uses ChatGPT to come up with content ideas, but if she asks ChatGPT to write a script and then reads it verbatim, she says, “You’re completely using it as a crutch and it’s a hindrance to your creativity.”
All three experts agree that it will be more difficult to parse out which parts (or percentages) of text were written and which parts were generated by AI. Goetze said that for now, “we should aim to be transparent” and be careful to avoid plagiarism and copyright infringement.
Another potential danger is becoming overly invested in and dependent on the responses of AI tools. “You don’t want to stay in a feedback loop with AI. You end up in a dark place,” Hafermalz says.
She recommends setting clear goals and intentions each time you use ChatGPT, and gradually increasing the stakes of your engagement so you always stay in control.
“It shouldn’t be a prison that locks you up. It should be a stepping stone. It should be a means to get out and do other things,” Hafermalz says. “Please use it in a place where you can verify it yourself in the real world.”
