How the AI works

Every question on Cognify is generated by a large language model (LLM) prompted against the official College Board Course and Exam Description (CED) for that course. The prompt includes the specific Learning Objectives, content standards, and exam structure for the unit being tested.

For MCQs, the model generates a stem with four answer choices that match the official format: one correct answer and three plausible distractors. Each question is tagged with its unit, topic, and difficulty level.

For FRQs, the model creates prompts that mirror past AP exam structure, along with a scoring rubric broken into individual point criteria. When students submit responses, a second LLM pass grades each criterion independently.

Cognify tracks your accuracy on every question, broken down by unit and topic. After each practice session, the system updates your proficiency profile -- a map of which areas you've mastered and which need work.

When generating your next test, the difficulty mix is adjusted based on your proficiency profile. If you're scoring 90%+ on a unit, the system increases difficulty. If you're below 60%, it provides more foundational questions to build understanding before advancing.

The goal is to keep you in the Zone of Proximal Development -- challenged enough to learn, but not so overwhelmed that you disengage. This is based on Vygotsky's educational psychology research and modern adaptive testing theory.

After each practice test, Cognify maps your percentage correct to the official scoring tables. For AP exams, this means the composite-to-AP-score conversion (1-5) published by the College Board. For the Digital SAT, this means the section scoring tables (200-800 per section, 400-1600 total).

Your predicted score is calculated using a weighted average of your recent performance, with more recent tests weighted more heavily. This reflects the reality that your current ability is more important than a test you took two weeks ago.

A trend line on your dashboard shows how your predicted score has changed over time. This helps you see the concrete impact of your study sessions and stay motivated.

Spaced repetition is one of the most well-researched learning techniques in cognitive psychology. The core insight: reviewing material at increasing intervals dramatically improves long-term retention compared to massed practice (cramming).

When you miss a question on Cognify, it automatically enters your review queue. The system uses a modified Leitner box algorithm to determine when each question should reappear. Questions you continue to miss appear more frequently; questions you answer correctly are spaced out further.

The review queue is surfaced on your dashboard with a daily count, so you always know how many items need review. Sessions are short (5-15 minutes) and designed to fit into existing study routines.