There’s an uncomfortable truth hiding inside the “AI will transform your grades” hype: used the wrong way, AI makes studying less effective, not more. The research on this is now solid enough to say it plainly students who let AI do their thinking retain less, even when their assignments look better.
But the same research points to something more encouraging. When AI is used to force retrieval, test understanding, and give feedback rather than to produce answers it amplifies the study techniques cognitive science has been recommending for decades. The difference isn’t the tool. It’s the workflow.
This guide covers the workflows that work: turning notes into practice questions, using the study modes built into ChatGPT and Gemini, grounding AI in your own course materials, and setting up spaced review. It also covers the failure mode what researchers call “metacognitive laziness” and how to stay out of it.
Start With the Science, Not the Software
Before any tool choice, one piece of learning science explains almost everything else in this article: reading a textbook cover to cover, highlighting passages, and rereading highlighted sections before an exam is one of the least effective study methods according to learning science research. Active recall, spaced repetition, and elaborative interrogation asking why and how are far more effective.
Active recall means retrieving information from memory instead of reviewing it. You cover the answer, try to produce it yourself, then check. That act of retrieval strengthens the memory far more than rereading does. The evidence behind it is striking: in Roediger and Karpicke’s 2006 experiments, students who self-tested forgot only 13% of a passage after two days, compared with 56% for students who reread. A 2013 review by Dunlosky and colleagues ranked self-testing among the most effective of 10 common study techniques, well above highlighting and summarising.
So why doesn’t everyone study this way? Because the biggest barrier to active recall isn’t the technique it’s the setup time. Writing good practice questions and flashcards for every lecture takes hours most students don’t have.
That’s the real case for AI in studying. Not shortcuts setup. AI eliminates that setup barrier completely. You can now generate recall questions, build spaced schedules, and test yourself on any material in minutes.
Workflow 1: Turn Your Notes Into a Quiz, Not a Summary
The instinct most students have is to paste notes into a chatbot and ask for a summary. Resist it a summary is passive review with extra steps. Ask to be tested instead.
A prompt pattern that works well: “Here are my notes on [topic]: [paste notes]. Generate 15 retrieval practice questions that test understanding, not just memorization.” Answer them from memory, out loud or on paper, before checking anything.
Then close the loop on your weak spots: on review days, retest yourself ask AI to quiz you again, but only on the topics you previously got wrong. This matters more than it sounds, because students instinctively focus review on material they already know because it feels good to get cards right. But research shows that the material you can’t recall is exactly what needs more retrieval practice.
Two other recall techniques pair naturally with a chatbot:
- The Feynman technique, with feedback. Explain a concept in simple terms as if teaching a beginner. If you can’t, you’ve found a gap. Type your explanation to the AI and ask it to point out what’s wrong or missing.
- Blurting, then checking. It involves reading a section of your notes, closing the notebook, and then ‘blurting’ out everything you remember onto a separate piece of paper. Paste your blurt into the AI alongside your notes and ask what you missed.
Workflow 2: Use the Study Modes, They Exist for a Reason
In 2025, the major AI companies all shipped tutor-style modes in quick succession: July 29: OpenAI launched “study mode” in ChatGPT. August 6: Google followed suit with “Guided Learning” in Gemini. August 14: Anthropic expanded its “Learning” mode to all users. The idea behind all of them is that instead of just handing you answers directly, study modes guide you through the thinking process via techniques like scaffolding, Socratic questioning, exercises, and so on.
The good news for anyone on a budget: you don’t need a paid plan. Study Mode is available to all ChatGPT users, including free-tier users. It makes ChatGPT Socratic: instead of giving you the answer, it asks guiding questions and gives hints. Gemini’s equivalent is available to all Gemini users, so you don’t have to pay for a subscription plan to gain access to it.
They behave differently in practice. Gemini feels more structured and visual, while ChatGPT’s Study Mode is conversational and flexible, letting you ask follow-up questions. One reviewer found ChatGPT easier for exploring topics deeply, but Gemini helps stay focused on systematic learning. Gemini also leans multimodal: the feature responds using images, diagrams, videos, and interactive quizzes to help users build and test themselves on their knowledge, rather than simply giving them the answer.
For dense reading assignments, one student tester shared a prompt worth stealing: upload the PDF and say “Walk me through this one section at a time. After each section, ask me a question to check I understood before continuing.” It turns passive skimming into something that sticks.
To get the most from either mode, start with context. Tell the AI what you want to learn, why, and what you already know. The model excels at adapting to your level and guiding you step by step.
Which one is “better”? Honestly, testers disagree one reviewer concluded ChatGPT takes the crown, with patient, milestone-based questions that actually teach, while another found that for a history or literature assignment, Gemini Guided Learning could take a primary source and walk through it paragraph by paragraph, asking what the reader thought each section meant before explaining exactly the kind of slow reading that builds actual understanding. Since both have solid free tiers, there is zero reason to limit yourself to one.
Workflow 3: Ground the AI in Your Actual Course Materials
General chatbots have a weakness for exam prep: they answer from the whole internet, and your exam comes from your professor’s slides. This is where source-grounded tools like Google’s NotebookLM earn their place. NotebookLM grounds every response in the documents you upload (lectures, papers, textbook chapters), so the AI cannot make things up.
That grounding is more than a convenience. Unlike general AI chatbots that can hallucinate or mix in information from unrelated sources, NotebookLM is grounded exclusively in the materials you upload. As one comparison put it, the split that matters: AI that grounds its answers in your specific sources protects your understanding; AI that hallucinates plausibly erodes it.
A simple exam-prep routine: upload the semester’s slides and readings into one notebook, then use it to generate practice questions per topic and quiz yourself again, retrieval first, answers second. Even a university revision guide now recommends this pattern: upload any notes or resources, then ask NotebookLM to test you on the topic.
Workflow 4: Add Spaced Repetition So It Sticks
Active recall tells you how to study; spacing tells you when. A single retrieval session is powerful. Multiple retrieval sessions spaced over time are dramatically more powerful. Active recall works best when combined with spaced repetition reviewing material at increasing intervals over days and weeks.
A practical schedule: test yourself within 24 hours of every lecture using your generated materials, then schedule spaced reviews at 1, 3, 7, and 14 days after initial learning.
On tooling, a hybrid approach beats purism. Anki is excellent for pure spaced repetition with flashcards. AI is better for generating deeper questions and adapting in real time to your weak areas. Ideally, use both AI for understanding and application questions, Anki for pure memorization. A common workflow: have a chatbot draft flashcards from your notes, edit them yourself (the editing is studying), then run them through Anki’s scheduler.
Left it too late? You’re not doomed: the ideal is to start from day one of the course. But realistically, starting 2-3 weeks before an exam still gives you enough time for 3-4 spaced review sessions.
The Trap: When AI Makes You Study Worse
Now the part most guides skip, and the part worth taking seriously.
A study published in the British Journal of Educational Technology gave it a memorable name. Researchers set up an experiment in which 117 students were asked to read several texts, write an essay on them and then revise the essay. The students were randomly assigned to one of four conditions: one group was allowed to use ChatGPT, the second could consult a human writing coach, a third was given a writing toolkit with checklists, and a fourth had no extra help.
The results cut both ways. Students in the ChatGPT group showed the most improvement in essay scores, even outperforming the human expert group suggesting that personalised, immediate feedback from AI can be effective. However, there was no significant difference in how well the ChatGPT group applied what they learnt to new situations. They received short-term gains without the long-term benefits. The ChatGPT group relied strongly on the AI support and showed relatively low metacognitive processing compared to the other groups, leading the authors to conclude that ChatGPT might promote “metacognitive laziness” where students refrain from engaging deeply in self-regulated learning processes.
There’s also a subtler mechanism: AI answers feel like understanding. Researchers call this the “illusion of fluency”, wherein students mistake the ease of processing highly polished, externally generated AI content for authentic internal comprehension. If studying with AI never feels hard, that’s not efficiency that’s a warning sign. The struggle to retrieve is the mechanism that builds memory.
The fix isn’t avoiding AI; it’s how the help is structured. One frequently cited finding: a Wharton School study found guardrail-based AI tutoring raised practice scores by 127 percent with no drop on unassisted exams, while unguided AI raised practice scores by less and still caused a 17 percent decline on that same unassisted test. Guardrails hints instead of answers, questions instead of solutions are exactly what study modes and the workflows above provide.
Three rules that keep you on the right side of the research:
- Retrieve before you ask. Attempt every problem, explanation, or blank-page recall before the AI sees it. Then use AI to check and correct.
- Ban copy-paste into your own work. Feedback on your draft builds skill; generated text replaces it.
- Test yourself without AI regularly. Your exam is unassisted. If your practice never is, you’re measuring the AI’s knowledge, not yours.
A Study Session Template You Can Use This Week
Putting it together, here’s what a 60-minute AI-assisted session looks like for one lecture’s material:
- Minutes 0–10: Blank-page blurt. Write everything you remember from the lecture, no notes, no AI.
- Minutes 10–20: Paste your blurt and your notes into a chatbot. Ask: “Compare these. What did I miss or get wrong?”
- Minutes 20–40: Ask for 10–15 retrieval questions that test understanding, not memorization. Answer from memory. Flag every miss.
- Minutes 40–55: Switch to study mode (ChatGPT or Gemini) and work through your flagged weak spots with guided questioning, not direct answers.
- Minutes 55–60: Schedule your reviews days 1, 3, 7, and 14 and note which topics need re-quizzing first.
Notice the pattern: your brain goes first in every step, and the AI responds to what your brain produced. That single design choice is what separates studying with AI from outsourcing to it.
The Takeaway
AI genuinely can make studying better but only in one direction. It removes the setup cost of the techniques that were always the most effective: self-testing, spaced review, explaining concepts, and working through hard material with guided questions. Used that way, it’s the cheapest personal tutor students have ever had. Used as an answer machine, the research is equally clear: assignments improve while learning quietly doesn’t. Keep the effort of retrieval for yourself, hand the busywork to the AI, and you get the best of both.
External Sources:
- British Journal of Educational Technology — “Beware of metacognitive laziness” (Fan et al., 2025)
- Educational Psychology Review — “Looking Beyond the Hype: Understanding the Effects of AI on Learning” (Springer)
- The Hechinger Report — coverage of student AI over-reliance research
- Roediger & Karpicke (2006) retrieval practice experiments; Dunlosky et al. (2013) study-technique review
- TechCrunch — Gemini Guided Learning launch coverage
- Birmingham City University — active recall revision guide
Read More on AI Global Class