How to ask an AI tutor better study questions
The AI tutor answers only from your material — no invented facts. The trick is asking questions you can actually study from.
The short answer
An AI tutor is only as useful as the question you ask it, and the questions that work are the ones shaped like exam questions. AuroraNote’s tutor answers from your uploaded material rather than the open web: each question retrieves the twelve most relevant passages from your own documents, and the answer is written from those, so it reflects your slides instead of a generic textbook. That grounding should change how you prompt it. Instead of "explain photosynthesis", which returns an encyclopedia entry you could have found anywhere, ask "what are three likely exam questions from section 2?", "explain the light phase as if I were fifteen", or "where do lecture 3 and the textbook disagree?". Group related documents into a Course to ask across all of them at once. Then end each session by asking the tutor to quiz you and grade your answers, which turns the conversation into retrieval practice.
The AI tutor answers only from your material — no invented facts from the open internet. That constraint is the feature: it behaves like a study partner who has actually read your slides, not a general chatbot guessing. But you still get out what you put in, and the quality of your questions decides the quality of your revision.
Here is how to prompt it so the answers are things you can actually study from.
What makes a study question weak or strong?
A vague prompt gets a vague answer. The useful distinction is not long versus short — it is whether the question specifies a level, a section, or a task. Exam-shaped prompts get answers you can revise from directly.
| Instead of | Ask | Why it is better |
|---|---|---|
| "Explain photosynthesis" | "Explain the light phase as if I were fifteen" | Naming a level forces an explanation instead of a definition |
| "Summarise chapter 4" | "What are three likely exam questions from section 2?" | Produces something you can be tested on, not more to read |
| "Is this important?" | "Where do lecture 3 and the textbook disagree?" | Uses the one thing this tutor has that a chatbot does not: your sources |
| "Did I understand this?" | "Quiz me with five questions and grade my answers" | Turns the session into retrieval practice rather than reading |
How does the tutor know what is in your documents?
Your documents are split into passages of roughly 450 tokens and indexed by meaning rather than by keyword. When you ask a question, the twelve passages closest in meaning to it are retrieved and handed to the model together with your question, and the answer is written from those passages.
Two consequences are worth knowing. First, the tutor can only tell you what is in the material you uploaded — if a topic was never in your slides, asking about it will not conjure it, which is the trade you make for answers that match your actual course. Second, the more precisely you name a section or a source, the better the retrieval step performs, because you are effectively helping it find the right twelve passages.
How do you chat across a whole course?
Group related documents into a Course and ask across all of them at once. Cross-referencing is where a tutor saves the most time: "where do these two files disagree?" or "which topics appear in both the lecture and the past paper?" surfaces connections you would spend an evening finding by hand.
The tutor also answers in the language you write in, independently of the language of the source. A lecture attended in one language can be revised in another without translating the document first.
Frequently asked questions
Does the AI tutor make things up?
It is grounded in your uploaded material: each question retrieves the twelve most relevant passages from your own documents and the answer is written from those, rather than from the open web. That does not make hallucination physically impossible — no language model guarantees that — but it means the answer is anchored to text you can go and check, and that a topic absent from your material will not be invented to fill the gap.
Can it use several documents at once?
Yes. Group documents into a Course and the tutor can answer across all of them, which is ideal for cross-referencing lectures, slides and past papers.
What are good prompts for exam prep?
Ask for likely exam questions from a named section, explanations pitched at a simpler level, comparisons between two of your sources, or to be quizzed and graded. The common thread is that each produces something you have to respond to, rather than more text to read.
Sources
- Roediger & Karpicke — Test-Enhanced Learning: Taking Memory Tests Improves Long-Term Retention — Psychological Science 17(3), 2006.