QPAssess

AI disclosure

How AI is used in your assessment

If you’re a student about to take a QPAssess quiz: an AI voice will ask you questions and listen to your answers. This page tells you exactly which AI systems hear you, what they decide, what they don’t decide, and what is kept afterwards. Last updated June 10, 2026.

What the AI sees, and which models

  • Transcription: your microphone audio is transcribed by OpenAI’s gpt-4o-mini-transcribe speech-to-text model, so your answers exist as text for your professor to read and for scoring.
  • The examiner you talk to: OpenAI’s gpt-4o model generates the examiner’s questions and follow-ups; its voice is OpenAI’s tts-1 text-to-speech. The examiner is given the question text, the rubric’s “what to look for”, and any persona or instructions your professor wrote — plus the live transcript of what you say.
  • Scoring: after you finish, the transcript and your professor’s rubric are sent to gpt-4o, which proposes a score per rubric criterion with a written justification, quoted evidence from the transcript, and a confidence estimate.
  • Professor-side tools: when a professor imports questions from a syllabus or generates a role-play scenario, gpt-4o processes the documents the professor provided. No student data is involved in those features.
  • What no AI model sees: webcam and screen recordings (on quizzes that use them) are never sent to any AI model. They exist solely for your professor to review.

What the AI decides — and what your professor decides

  • The AI proposes: per-criterion scores, justifications, transcript quotes as evidence, and a confidence level.
  • Your professor decides: they review each response, can change any score (every override is logged), and nothing counts as a grade until they mark it reviewed.
  • When grades sync to Canvas, only professor-reviewed scores are sent — never raw AI scores — and only when the professor triggers the sync.
  • Guardrails on the proposal: AI scores are clamped to the rubric’s point ranges, evidence quotes are checked word-for-word against the transcript (made-up quotes are discarded), and anything the system can’t verify is flagged for human review instead of being scored.
  • If your professor turns on instant feedback, you’ll see short AI-written study feedback right after the quiz — it never includes or reveals a score.

What is stored

  • The transcript of your conversation, the audio recording (plus webcam/screen video only on quizzes that use them), the AI’s proposed scores with evidence excerpts, and your professor’s final scores and notes.
  • Your name and email as you entered them, and when you started and submitted.

What is not stored or used for training

  • We do not train AI models on your data, and per OpenAI’s published API data policy, data sent to the OpenAI API is not used to train OpenAI’s models unless the customer explicitly opts in — QuestionPunk has not opted in. OpenAI retains API data for up to 30 days for abuse monitoring, then deletes it unless legally required to keep it.
  • We do not run facial recognition, voice-print matching, or any biometric identification on recordings.

Retention

Honestly: transcripts, scores, and recordings are kept until your instructor or institution asks us to delete them — automatic deletion schedules don’t exist yet.

Want the things the AI examiner is known to get wrong? See Known limitations. For the broader data-handling picture, see the Trust center.