AI study assistants: how to use them to learn, honestly

AI chatbots can explain a hard concept at two in the morning, quiz you before an exam and point out the gaps in your reasoning. They can also do your thinking for you, which is where the trouble starts. Below are the rules you need to check first, the study features the main assistants now have, the uses that help you learn and the ones that quietly undermine it, and how to credit AI when you use it in your work.

Check the rules before you start

There is no single rule on AI at college. As Cornell University’s Learning Strategies Center puts it, every class and every instructor has different expectations about whether, when and how students may use AI, and those expectations should be in the syllabus. The same tool can be encouraged in one course and banned in the next. If you are not sure whether a particular use is allowed, ask your instructor before you do it. The responsibility is yours, not theirs.

Check, too, which tool your university supports. Cornell recommends Microsoft Copilot to its students because, when they sign in with their university ID, their prompts and answers are not used to train the underlying models. Ask what your own school provides before you sign up for anything.

The study features built into the main assistants

The big AI companies have added modes designed for learning rather than getting answers.

ChatGPT’s study mode, launched by OpenAI in July 2025 and opened by choosing “Study and learn” from the tools menu, works through problems step by step instead of handing over the answer. It asks guiding questions, gives hints, prompts you to reflect, and checks what you have understood with quizzes and open-ended questions. You can turn it on and off within a conversation. OpenAI itself notes that it can behave inconsistently and make mistakes.

Anthropic’s Claude has a learning mode, introduced with Claude for Education in April 2025. It is built to guide your reasoning rather than answer for you, asking questions such as “How would you approach this problem?” and “What evidence supports your conclusion?”

Google’s Gemini Notebook works from material you give it. You upload your own sources, such as PDFs, websites, YouTube videos, audio, and Google Docs or Slides, and its answers are grounded in those sources, with in-line citations showing where each point came from. It can turn your sources into study guides, audio overviews and mind maps.

Microsoft Copilot works inside Microsoft 365 apps.

Uses that help you learn

Cornell’s center makes a distinction worth keeping in mind: AI is very good at routine tasks and at pulling information together, mostly accurately, but it cannot create new knowledge, and when students use it to outsource their thinking it can do more harm than good.

Used as a tutor, it can help. Having an AI quiz you is, in one Cornell student’s words, like a virtual study buddy. Asking it to explain a concept in a different way can help you understand an idea the textbook did not make clear. Checking your own worked answer and asking where you went wrong is a good use; the center rates getting help after you have tried a problem yourself as the most useful kind of help.

What the research shows so far

Two controlled studies published in 2025 show both sides. In a field experiment with nearly a thousand high school math students, researchers led by Hamsa Bastani gave some students GPT-4 to use during practice. With a standard chatbot, practice grades rose 48%, but when the AI was taken away, those students did 17% worse on their own than students who had never had it: they had used it as a crutch. A second version, built with safeguards to tutor rather than answer, raised practice grades 127% and largely avoided that drop.

At Harvard, a team led by Greg Kestin ran a randomized experiment in a large undergraduate physics course with 194 students. Students who worked through lessons with an AI tutor built on the same teaching principles as the course’s own active-learning classes learned significantly more in less time than they did in class, and said they felt more engaged and motivated.

The lesson for students is the same one Cornell draws: an assistant set up to make you think can help you learn, and one that hands you answers can leave you worse off once it is gone.

Uses that undermine learning

The same center lists the pitfalls. If AI writes your summaries and flashcards, you lose much of their value, because the learning comes from making them, from deciding what matters and connecting it to what you already know. If AI solves problems for you, its explanations can convince you that you understand a method you could not use on your own, and you will find out in the exam. Being stuck and frustrated is part of learning.

AI also gets things wrong. You need to know enough about the topic to notice when it does, which is hard when you are learning the topic for the first time. Study plans it writes can look polished but turn out generic. And it is not always faster: Cornell’s center warns that AI may not produce good work faster than you can. In a 2025 study by the research group METR, experienced open-source software developers working on real tasks in projects they knew well took 19% longer when they were allowed to use AI tools, even though they expected AI to speed them up.

Credit it when you use it

When AI contributes to work you hand in, and your instructor allows it, say so. The APA and MLA style guides both explain how.

APA Style asks you to describe how you used the tool, in the method section of a research paper or the introduction of an essay, and to give the prompt you used along with the relevant text it produced. Its reference format names the company, year, tool and version, for example: OpenAI. (2023). ChatGPT (Mar 14 version) [Large language model]. Long responses can go in an appendix.

MLA Style says to cite a generative AI tool whenever you paraphrase, quote or include any content it created, and to acknowledge other uses, such as editing your prose or translating words, in a note or in your text. It also warns you to check any sources an AI cites, because they need to be verified before you rely on them.

A simple test

Before you open a chatbot, ask what you are trying to get out of the task. If the point of the assignment is to build a skill, such as solving a type of problem, writing an argument or planning a project, then handing that work to AI takes away the very thing you are there to learn. Use AI to check, question and explain. Do the thinking yourself, and never submit AI-written work as your own.