A useful AI study question usually needs less personal data

Students can ask for feedback while keeping classmates’ work, research participants and private records out of the prompt.

A student asking for help with a concept rarely needs to upload an entire folder of personal information. A short fictional example can often explain the difficulty. Reducing the input makes the question clearer and avoids exposing classmates, teachers or research participants. Privacy begins with deciding what the task actually requires.

UNESCO’s [2023 guidance for education and research](https://www.unesco.org/en/articles/guidance-generative-ai-education-and-research) identifies data and governance concerns. Oxford’s [research AI policy](https://www.ox.ac.uk/research/support/governance-and-committees/research-policies/policy-for-using-generative-ai-in) addresses responsibilities for research material, while its [student AI questions](https://www.ox.ac.uk/students/life/it/genai-tools/genai-faqs-for-students) provide institutional context. These primary sources support careful use. They do not establish that every education-branded account permits every kind of upload.

Before submitting material, check the course rule and the institution’s approved tool guidance. A student licence may have different settings from a personal account with a similar name. The relevant question is whether the specific tool and data type are authorised for the task. Younger learners also need the school’s age, supervision and safeguarding arrangements.

An original illustrative student wants feedback on the organisation of a group project. The shared document contains classmates’ names, contact details and private comments. Instead of uploading it, the student writes a fictional outline of the structure and asks where the argument is unclear. This achieves the learning purpose without sharing the other students’ information.

Removing names is not always enough. A detailed account of a rare event, a distinctive quotation or a small participant group can still identify someone. Research students should follow their approved ethics and data-handling process before using any participant material. Consent to take part in a project does not automatically imply permission to submit information to a new external service.

A survey-analysis exercise can use prepared sample data. Ask the assistant to explain a method on fictional rows, then perform the real analysis within the authorised environment. Do not ask it to invent missing participant answers or present synthetic data as collected evidence. Learning how an analysis works is different from changing the research record.

Documents can contain hidden information too. Comments, revision history, embedded images and file properties may reveal details absent from the visible page. A screenshot can include a name, email or notification. Review what you plan to share rather than assuming the main text is the entire input. Use the approved institutional process when sensitive material is involved.

The assistant’s output also needs care. It may infer a person’s circumstances, repeat a private detail or generate a misleading account. Do not forward the response to a public group simply because it looks useful. Check what it contains and whether sharing is permitted. A summary can be sensitive even when it is shorter than the original.

Retention and sharing settings require an authoritative explanation. Read the institution’s guidance for the actual service or contact its support team. Do not assume that deleting a chat settles every data-handling question. If you accidentally submitted inappropriate material, use the institution’s incident or support route promptly and explain what happened without spreading the material further.

Keep learning records proportionate. A course may ask you to document AI assistance, but that record should not become an unnecessary copy of personal data. Describe the type of task and retain permitted excerpts or drafts under the course’s instructions. Protect credentials, account tokens and access links from screenshots, assignments and shared notes.

The potential benefit is useful feedback with a smaller privacy burden. The failure is treating convenience as permission to upload information about other people. A clear study question, a fictional example and an approved tool can often provide the same learning support. When the task genuinely needs sensitive data, obtain the proper institutional guidance before proceeding.

A final check can be phrased in ordinary language: whose information is this, why does the tool need it and am I authorised to share it here? If you cannot answer those questions, rewrite the question with less information or seek help. That keeps AI assistance connected to responsible study rather than turning a learning exercise into an avoidable disclosure.