AI performance analysis

Track these across attempts, not within one test.

• Accuracy per subject and per chapter
• Attempt rate and unattempted count
• Average time per question by difficulty band
• Negative marks caused by guessing versus by misconception
• Repeat-error rate — the same concept failing twice

A single low score after a syllabus jump is normal. Three flat attempts with the same chapters failing is a plateau, and it needs a change of method, not more hours.

If you ask AI for the answer, you get the answer and lose the skill. Escalate your prompts in this order and your accuracy in the real exam improves instead of your dependence.

• Weak: "Give me the answer."
• Better: "Give me one hint, not the solution."
• Best: "Here is my attempt and my reasoning — tell me exactly where it breaks."
• Target loop: Think → Attempt → Get feedback → Correct → Practise → Master.
• Anti-pattern: Ask AI → copy answer → move on.

Your Know Your Level report and Prepkunj chapter test results together give a chapter-level accuracy view. Bring them to a free counselling call and a mentor will turn them into a plan.

Every benefit below has a matching failure mode. Name them out loud with your student so they are managed rather than discovered in February.

• Hallucination: convincing but wrong facts, formulae or values — verify against NCERT and official syllabi.
• Overdependence: problem-solving stamina drops when AI does the first step every time.
• Academic integrity: work submitted without understanding teaches nothing and shows up in tests.
• Data privacy: educational AI processes sensitive student data; UNESCO's guidance stresses human-centred, age-appropriate, privacy-respecting use.
• Bias: models reproduce the biases in their training data and design.
• Reduced human contact: teachers, mentors, peers and family still drive most of a student's progress.