AI can make studying more effective or it can make it feel more effective, and those are different things. A model that summarises your lecture into a neat page will make you feel prepared. A model that asks you questions you can't answer yet will actually prepare you.
This guide is about the second kind. Here's which model suits each part of exam prep, and how to use it so that you're the one doing the remembering.
The short version
| Job | Good pick | Why |
|---|---|---|
| Turning notes into flashcards | ChatGPT (GPT-6.1 Sol) | Clean, structured output in the format you ask for |
| Practice questions that test understanding | Claude (Sonnet 5.5) | Good at questions about why, not just what |
| Summarising a long lecture recording or slide deck | Gemini | Takes in audio, slides and PDFs together |
| Quizzing you, one question at a time | Claude or ChatGPT | Both handle back-and-forth well |
| Lots of quick checks on a budget | Gemini 3.8 Flash or DeepSeek V4.1 Flash | Cheap per message |
| Questions about current topics | Perplexity | Answers with sources you can open |
Lecture notes: Gemini for the raw material
Lectures come as slides, recordings, handouts and your own scrappy notes. Gemini is built for exactly that mix: attach a long PDF, a slide deck or an audio file and ask across all of it. On LazySusan, Gemini models accept prompts of up to 1M tokens.
The slide deck is from this week's lecture and the text file is my notes. Tell me which points on the slides are missing from my notes, and which things I wrote down that aren't on the slides. Don't summarise the whole lecture.
That's more useful than a summary, because it shows you what you missed.
Flashcards: ChatGPT for format
Flashcards are a formatting job, and GPT-6.1 Sol is good at producing exactly the structure you ask for.
Make 20 flashcards from these notes as a two-column table: question on the left, answer on the right. One fact per card. Questions should require recall, not recognition, so no yes/no or multiple choice. Use my wording for key terms.
Then check every card against your notes. A wrong flashcard is worse than none, because you'll learn the mistake.
Make the cards yourself when you can. Writing the question is part of learning the answer. Use AI to fill gaps or to turn a long reading into a first draft you then edit.
Practice questions: Claude for "why" questions
Recall cards test what. Exams usually test why and how. Claude Sonnet 5.5 is good at writing questions that make you explain a mechanism, apply an idea to a new case, or spot where a simple rule breaks down.
Based on these notes, write 8 exam-style questions: 3 that test definitions, 3 that ask me to apply a concept to a new example, and 2 that ask me to compare two ideas. Don't give the answers yet.
Answer them on paper, then:
Here are my answers. Mark each one as an examiner would, tell me what a full-marks answer would include that mine doesn't, and don't be generous.
"Don't be generous" helps. Models lean towards encouragement.
Quiz mode: one question at a time
The most effective way to study with AI is to let it question you, not the other way round.
Quiz me on [topic], one question at a time. Wait for my answer before the next question. If I get one wrong, explain briefly and come back to that idea two questions later. Get harder as I get things right.
Both Claude and ChatGPT do this well. If you get stuck, ask for a hint rather than the answer.
Mixing models
Different models explain the same idea differently, and sometimes the second explanation is the one that clicks. On LazySusan, Multi-Chat sends one question to several models and shows the answers side by side. Use it for the concept you're stuck on, not for every flashcard: each column is billed as a normal message. It's built for desktop and larger screens.
For fact-heavy subjects, cheaper models such as Gemini 3.8 Flash or DeepSeek V4.1 Flash are fine for quick checks. Save Claude Opus 5.5 or GPT-6 Astra for the topic you genuinely don't understand.
Where AI study help goes wrong
- Summaries replace reading. A summary is someone else's choice of what matters. Read the original for anything you'll be examined on.
- Wrong answers sound right. Models make mistakes in a confident tone. Check against your notes, textbook or course materials.
- It's too easy. If every quiz question feels easy, ask for harder ones. Struggle is the point.
- It doesn't know your course. Your instructor's emphasis, notation and past papers matter more than a general explanation. Give the model your materials and say "use only these".
A week-before-the-exam plan
- Day 1: Use Gemini to compare your notes against the slides for each topic. Fill the gaps from the source.
- Day 2: Make flashcards (yourself, or ChatGPT for a first draft you edit).
- Day 3–5: Daily quiz sessions with Claude or ChatGPT. Note every question you miss.
- Day 6: Ask for exam-style questions on only the topics you missed. Answer on paper, then get them marked.
- Day 7: Past papers without AI. That's the real test.
The verdict
There isn't one best AI for studying, there's a best one for each step: Gemini for working through lecture material, ChatGPT for flashcards and tidy revision sheets, Claude for questions that test understanding and for honest marking. Any of them will quiz you well.
Having all of them in one place is easier than juggling apps. LazySusan includes them in every plan from $8 a month, with a Student plan at $99 a year for .edu emails. See pricing for details. If your prompts aren't getting what you want, Prompt School covers the basics.
Use it honestly
- Check your course's AI policy, especially for take-home exams and graded quizzes, where AI is often not allowed at all.
- Never use AI during an exam or assessment unless it's explicitly permitted.
- Don't submit AI work as your own where it isn't allowed. Study help and graded work are different things.
- Check what it tells you. If a model's answer disagrees with your course materials, go with the course and ask your instructor.