Use AI to create opportunities to think, then check what you can do without it. Follow your course’s rules for permitted use.
A useful AI study session starts with a learning goal and ends with something you can explain or solve independently. Try the task yourself, ask for a targeted hint, check the explanation against course material, then attempt a fresh problem without help. The workflow below is a suggested study structure, not a proven recipe for higher grades.
UNESCO’s guidance on generative AI in education places human agency and educational suitability at the centre of tool use. Our practical interpretation is to keep the learner making the important attempts and decisions. A completed answer is a poor record of learning if you cannot explain how it was produced.
Check the rules before choosing the task
Read the assignment instructions and institutional policy. Permitted brainstorming, tutoring and assessed work may have different boundaries. When instructions are unclear, ask the teacher which uses are allowed and what disclosure is required. A tool being available does not establish permission to use it for an assessment.
Use material you are authorised to share. Do not upload another student’s work or personal details merely to get a more specific answer. Our privacy checklist explains how to reduce unnecessary information. Students who do not meet a service’s age requirements should use the arrangements provided by their school or responsible adult.
Begin with an attempt you can inspect
Choose a narrow objective, such as explaining percentage change or identifying the argument in a short passage. Write what you already understand and attempt one example. Mark the step where you became uncertain. This gives a tutor, human or AI, a specific problem to address.
Compare two requests. “Explain everything about percentages” invites a long overview. “I calculated the change from 40 to 50 as 20%; help me identify which starting value belongs in the denominator” exposes a particular confusion. You still need to check the guidance, but you now know what you are checking.
Ask for a hint before a full solution
Try this reusable instruction: “Act as a practice partner. Ask one question about my next step. Wait for my attempt before offering another hint. Do not produce an assessed submission for me.” This is a suggested prompt, not a guarantee that a system will follow every boundary.
If the answer gives everything away, set it aside and choose a new example. The purpose of the session is to practise decisions you will need to make yourself. Keep a short note of the misunderstanding that changed, rather than copying an entire generated explanation into your study notes.
Check explanations against trusted material
NIST’s generative AI risk profile identifies confident false outputs as a risk. For studying, use the assigned text, worked examples or teacher’s explanation to resolve disputed claims. A chatbot inventing a convincing reference can send you away from the subject instead of helping you understand it.
For a hypothetical percentage exercise, going from 40 to 50 is an increase of 10 divided by 40, or 25%. Returning from 50 to 40 is a decrease of 10 divided by 50, or 20%. Work both calculations yourself. The different starting values explain the different percentages; memorising the two answers misses that distinction.
Finish with an independent check
Close the chat and write the idea in your own words. Solve a fresh example and explain why you chose the method. Then compare it with trusted material. If you cannot continue without the assistant, identify the missing step and practise that step next.
A simple session record can contain four lines: my objective; my first mistake; the explanation I checked; what I could do alone. This is our suggested reflection template. It makes the session easier to discuss with a teacher and separates time spent using a tool from evidence of understanding.
To choose broader development goals, read skills to learn in the AI age. To inspect generated references, follow our answer-checking workflow.