
You open an AI assistant the night before a test and ask it to explain the chapter. The answer is clear, and you feel ready. Two days later, the details are gone.
Many students know this feeling, and finals season is close. A 2026 Lumina Foundation–Gallup survey of about 3,800 US college students found that 57% use AI for schoolwork daily or weekly. About 21% use it every day. So the better question now is how to study with AI so the knowledge stays in your head.
This guide covers what the research shows and what to do tonight.
How to study with AI: the quick answer
To study with AI without losing your memory, let the AI ask the questions and answer them yourself. Use it as a tutor that quizzes you, checks your answers and schedules reviews. Avoid asking it for finished answers to read. Memory grows when you pull facts out of your own head.
This guide calls that approach Tutor Mode. Asking the AI for answers to read is Answer Mode.
Why does this matter right now?
Student use of AI has grown faster than the guidance that comes with it. As a result, most students work out their own habits, often the night before a test.
The US numbers show the gap. In the Lumina Foundation–Gallup study, 48% of students at schools that discourage AI still used it weekly. So did 27% of students at schools that prohibit it.
The UK shows a similar pattern. The 2026 Higher Education Policy Institute (HEPI) survey questioned 1,054 full-time undergraduates in December 2025. It found that 95% use AI in some form. HEPI also described a student body divided on whether AI helps or harms learning.
Do not compare the two figures directly. Lumina–Gallup measured how often students use AI. HEPI measured whether they use it at all.
Some readers of our post on why you forget what you study asked about AI. They wanted to know whether it makes forgetting better or worse. This post answers that question.
Why does the same AI help some students and hurt others?
The same tool gives opposite results because it can do two different jobs. It can do the thinking for you, or it can make you do the thinking. Your memory follows whoever does the work.
Picture two students with the same biology chapter. Student A types, “Explain this chapter.” She reads a neat summary and feels ready. Student B types, “Quiz me on this chapter, one question at a time, and wait for my answer.” He gets some answers wrong and sees his gaps right away.
Both students spent 30 minutes with AI. Only Student B practiced pulling facts out of memory.
What most students miss about fluency
Psychologists use the word “fluency” for the feeling that you understand something because it reads smoothly. A clear AI explanation creates fluency fast. But fluency is not memory. You can follow an explanation perfectly and still fail to rebuild it a week later.
Students also judge this poorly. In a 2011 study in Science, Jeffrey Karpicke and Janell Blunt compared two ways of studying science texts. Most students expected concept mapping to work better. In fact, practicing retrieval produced more learning.
A related risk is “cognitive offloading,” which means handing mental work to a tool. A calculator does this for arithmetic. An AI assistant can do it for thinking. In an exam hall, though, no assistant is available.
I saw this in my own classes. In 14 years as a trainer, I taught more than 30,000 students of different abilities. I used neuroscience-based activities to sharpen memory. One activity was simple. Two students tossed a balloon to each other and repeated the lesson aloud while they played.
I used it with a class of trainees preparing for a job role. They found it hard to memorize instructions and keep pace with the other students. The activity helped them. Since then, hundreds of students have improved their memory with this technique. Saying a lesson aloud during a game asks the brain to produce it, not just take it in.
I tried the activity myself before I brought it into the classroom. Through it, I learned counting in a South Indian language, which I could not have learned otherwise. After 12 years, I still remember it.
This is one trainer’s experience, not a controlled study. Still, it points the same way as the research above. Memory sticks when you work to produce the material.
What does the research show about studying with AI?
The evidence points both ways. AI helps when it is built to make students think. It seems to hurt when it does the thinking for them. Three lines of research show the range.
Can an AI tutor beat a classroom?
In one trial, it did. In a 2025 Scientific Reports study, Harvard physics instructors tested a custom AI tutor against an in-class active learning lesson. The trial ran in fall 2023 with 194 eligible students in an introductory physics course. Each student tried both formats in consecutive weeks. The AI lessons took place at home.
The AI group’s median learning gains were more than double those of the classroom group. Students also finished faster. The median time with the tutor was 49 minutes, against about 60 minutes of learning time in class. Students also rated the tutor higher for engagement and motivation.
The design matters here. The tutor guided students through each problem in order. The team also gave it written step-by-step solutions to improve accuracy, and building the platform took several months. The authors note that chatbots are generally built to be helpful, not to promote learning.
Does AI-made quizzing beat rereading?
A 2026 paper in the Cognitive Science Society proceedings tested this. In the study, students either reread texts or answered AI-generated questions with AI feedback. On delayed tests, the quiz groups answered more questions correctly than the spreader. That result fits decades of work on retrieval practice. Retrieval practice means pulling information from memory instead of looking at it again.
Still, one caution applies. A 2024 study in the Journal of Experimental Psychology: Applied found that testing may not help when students write their own questions. Those questions often missed material on the later test. So AI-written questions might avoid that problem, but that study did not test AI.
Does using ChatGPT make you forget what you learn?
It may, when it replaces your own effort. A 2025 MIT Media Lab preprint followed 54 adults writing essays with a chatbot, a search engine or no tool. EEG recordings showed the weakest brain connectivity in the chatbot group. Those writers also struggled to quote their own essays.
Only 18 people finished the final session, and the task was essay writing. So the study is an early warning and says little about exam memory.
Answer Mode vs Tutor Mode: what is the difference?
In Answer Mode, you ask and the AI tells you. In Tutor Mode, the AI asks and you answer first. The tool stays the same, and only your instruction changes.
| Answer Mode | Tutor Mode | |
| Who starts | You ask a question | The AI asks you a question |
| Who does the thinking | The AI | You, with the AI checking |
| What you do | Read and copy | Recall, answer and correct |
| How it feels | Smooth and easy | Harder, with some wrong answers |
| Memory after a delay | Likely weaker, like rereading | Likely stronger, like retrieval practice |
| Best use | Clearing up one stuck point | Reviewing what you already studied |
Think of a map app. If it speaks every turn, you can reach the place and still not know the way. If you try the route first and check the map only when lost, you learn the route.
Many students confuse Tutor Mode with a tool that refuses to explain. A good tutor still explains. The explanation just comes after your attempt.
How to use AI as a tutor, not an answer machine: five prompts
Paste your own notes into the AI and tell it how to behave. Ask for one question at a time, and answer before it replies. Here is how to study with AI tonight, using five prompts.
Use your own notes or textbook pages, and tell the AI to rely only on them. That keeps the questions close to what your exam covers.
1. Quiz me. “Here are my notes: [paste]. Ask me one question at a time using only these notes. Wait for my answer. Do not give hints unless I ask.” These forces recall, and it resembles the setup in the conference study above.
2. Check my answer. “Compare my answer with my notes. Tell me what I got right, what I missed and one fix. Do not rewrite my answer.” You get feedback without handing over the thinking.
3. Explain it back. “I will explain [topic] from memory. Point out gaps and errors, then ask me two follow-up questions.” A blank-page explanation shows what you only recognize.
4. Mix it up. “Pick two ideas from my notes that look alike. Ask me to tell them apart and give an example of each.” Similar ideas cause many exam errors, so this targets them.
5. Save the misses. “List the questions I missed or hesitated on. In my next session, I will paste them back. Ask them again with different wording.” Copy the list into your notes yourself. The assistant may not remember today’s chat.
A short session like this shows quickly what you know and what you only recognize. If you are new to AI assistants, our beginner’s guide to using an AI assistant covers the basics.
How do I add AI to the 7-day forgetting-curve plan?
Use AI for your review sessions, not for the first read. Each review becomes a quiz you answer from memory, with the AI checking.
The forgetting curve describes how memory fades quickly after learning unless you review. In our 7-day plan, each day has a recall step. Here is how to study with AI inside it. Keep each AI session within that days’ time block.
- Day 1: Learn the new topic and do your blank-page recall on paper first. Then paste your notes and run Prompt 3 on the same topic to find gaps.
- Day 2: Recall Day 1 from memory first. Then run Prompt 1 on your Day 1 notes, followed by Prompt 2.
- Day 3: Learn the next topic. Then run a quick round of Prompt 1 on Day 1.
- Day 4: Run Prompt 1 on the Day 2 and Day 3 topics. Add Prompt 4 for ideas that look alike.
- Day 5: Ask for mixed practice questions on all topics so far. Save your misses with Prompt 5.
- Day 6: Paste in your list of misses and ask the AI to re-ask only those, with different wording.
- Day 7: Do a full recall of the week. Draw a mind map from memory, then compare it with your notes. Our guide to mind mapping tools can help. Then rest.
The Harvard authors call spacing a well-established way to improve retention. They also say future studies should combine it with AI tutoring. So treat this as a sensible combination, not a tested method. The base plan is itself an example, not a research-proven schedule.
What mistakes make AI study sessions backfire?
Most mistakes share one cause: skipping your own recall. Watch for these five.
- Copying answers. Pasting AI answers into your notes feels productive. It trains you to recognize answers, not produce them.
- Trusting wrong answers. Chatbots can sound sure when they are wrong. The Harvard authors noted this flaw and gave their tutor written solutions. Check key facts against your textbook.
- Asking for the answer too early. If you ask before trying, you lose the recall step. Try first, even if you guess.
- Rereading AI summaries. The conference study found quizzing beat rereading on delayed tests. Rereading a polished summary is still rereading.
- Using one long session. Short sessions spread over several days fit the spacing idea better.
What should parents and teachers know?
Judge an AI study session by who did the thinking. Ask to see the questions the student answered, not the summary the AI wrote.
For parents, this means a simple habit. Ask your child to show the quiz. Make Tutor Mode prompts the default, and keep direct answers for one stuck point. Rules alone may not change behavior. In the Lumina–Gallup study, students kept using AI at schools that discourage or prohibit it.
For teachers, the Harvard authors suggest using AI tutors to teach introductory material before class. Class time then goes to problem-solving and group work. They advise against replacing in-person teaching, and they warn against uses where students may skip the thinking. Teachers who want to match this to younger learners can start with our guide to how Gen Z learns best.
The bigger picture
The question is shifting from “should students use AI?” to “what job do we give the AI?” As a result, schools that teach Tutor Mode may protect memory better than schools that only set rules. That is a reasoned view, but no study has proved it yet.
What is still uncertain about studying with AI?
Several limits matter. Here is what the current studies cannot tell us.
- The Harvard trial covered one course over two weeks. Students took post-tests after each lesson, and the paper does not report a later retention test.
- The tutor was custom-built with expert-written prompts. A general chatbot with a short instruction may not match those results.
- The authors say structured AI tutoring may not always beat in-class learning, for example on tasks that need complex synthesis.
- The MIT study was small, and it is a preprint. Its task was essay writing, so it may not transfer to studying for exams.
- Surveys show how many students use AI. They cannot show whether AI raises or lowers grades.
- None of these studies compares Tutor Mode and Answer Mode for exam prep over months.
What could happen next
Two developments are worth watching. First, researchers may test AI tutoring together with spacing, as the Harvard authors propose. Second, schools are supplying more tools. In the HEPI 2026 survey, 38% of UK students said their institution provides generative AI tools. That figure was 23% in 2025 and 9% in 2024.
If institutions supply tools, the way students use them will matter even more. Long-term studies would show whether Tutor Mode protects memory.
The bottom line
How to study with AI comes down to one question: who does the remembering? Let the AI ask, answer from memory, and let it check you. Reading its answers can feel like learning, but the evidence favors recall.
Tonight, pick one chapter, paste your notes and run Prompt 1. Save your misses for Day 1. The research is still young, so watch for long-term results before trusting any big claim.
Get new Chatterlane articles in your inbox. Subscribe to the newsletter
Frequently Asked Questions
Is it okay to use AI to study?
Yes, if it makes you think. Use it to quiz you and check your answers. Also follow your school’s rules, since policies differ.
Does using ChatGPT make you forget what you learn?
It can when it does your thinking. A small MIT preprint linked chatbot use in essay writing to weaker recall of one’s own work. More research is needed.
How do I use AI as a tutor, not an answer machine?
Paste your notes and tell it to ask one question at a time and wait for you. Ask for feedback, not rewrites. Prompts 1 and 2 above do this.
What are the best AI prompts for retrieval practice?
The best ones make you answer first. Quiz me, check my answer and explain it back are good starting points. Prompts 1 to 3 above cover them.
Can AI help students learn faster?
It can. In one Harvard physics trial, students using a carefully built AI tutor learned more in less time than in an active-learning class. That was one course, so results may differ elsewhere.
About the author: Rajesh Jain is a content strategist and writer. He spent 14 years as a trainer, teaching more than 30,000 students of different abilities and using neuroscience-based activities to help them improve their memory.


Leave a Reply