By the Feedback Flows education team
Questions to Ask About Student Data and AI
Several institutions are reviewing contracts around student work and AI development. If you buy marking, feedback or integrity tools, these questions help you compare vendors on substance rather than slogans.
1) Who owns uploaded student work?
Ask for a clear statement that the institution (or the uploading teacher) retains ownership of submissions. Marketing pages often imply this; your DPA and terms should say it outright.
2) Can the provider use work to train or build AI?
Separate two claims: “we do not sell student data” and “we do not use submissions to train or improve models.” Ask whether learner work can be used for product development, model fine-tuning, evaluation sets or shared corpora. If the answer is “only anonymised,” ask how anonymisation is defined and whether it is optional or contractual.
3) Can those terms change mid-contract?
Ask how notice works if the provider revises its EULA, privacy policy or AI training terms. Useful follow-ups: opt-out rights, termination for convenience if terms move against you, and whether historic submissions stay under the old rules.
4) Where is content stored, and who processes it?
Request a live sub-processor list with purpose and location. Distinguish file storage from AI processing. A UK-hosted file store does not automatically mean UK-only AI inference. Ask which transfer mechanisms cover overseas processors.
5) What are the retention periods?
- How long do original submission files stay on disk?
- How long do marks, feedback and detection runs remain?
- What happens to video or audio media?
- Is there an automated purge, or only deletion when a user asks?
6) How do you export or delete on exit?
Ask for the practical path: export formats, who can trigger deletion, and what the DPA says about return or destruction on termination. “Contact support” is weaker than a documented self-serve export plus a contractual delete-or-return clause.
7) Does integrity tooling produce evidence, or only a score?
A single AI percentage is a weak basis for a student conversation. Prefer tools that show quoted evidence, compare against a learner’s own writing history, check references, and leave a recorded human outcome. See our companion guide on evidence over AI detection scores.
How Feedback Flows answers these
We publish plain-English commitments on how we handle learner work: you own the work; processing is purpose-limited; we do not use learner work to train or build AI models; our AI provider does not train on API content; retention is published; sub-processors stay in sync with the legal pages; and you can export or delete at any time. The Privacy Policy and DPA remain the binding terms.
Universities and colleges can also start from our for universities page, or ask for a security pack via contact.