Professors & teaching teams · Workflow 08
Turn course feedback into teaching improvements without identifying students.
Create a teaching-improvement agenda from manually grouped feedback themes and verified counts, without sharing individual student records.
The moment this is useful
The task on your desk
After a research-methods module, a professor wants to plan next term’s improvements. Students have commented on the timing of assignment guidance and the pace of worked examples. Some comments also mention names, extensions, disability adjustments, or distinctive project topics. The task is to improve the module, not evaluate individual students.
Keep the facts the question depends on.
Keep for the research
- Specific teaching issues, such as late assignment guidance or worked examples moving too quickly.
- Contrasting views and uncertainty about whether a theme is widespread.
- Verified aggregate counts only when appropriate for the group and permitted by university policy.
Remove or replace
- Names, email addresses, student IDs, grades, and links to individual submissions.
- Health information, disability adjustments, extension reasons, and other personal circumstances.
- Unique dissertation topics, distinctive quotations, and small-group details that identify a student.
A worked prompt example
Fictional details · Edited by hand after reviewBefore review
Fictional feedback: Taylor Example, student ID DEMO-014, mentions a personal extension and asks for the rubric earlier. Another student says the worked statistical example moved too fast; a third says the pace was right. Prepare next term’s improvement plan.
Reviewed prompt
Draft a teaching-improvement agenda for a research-methods module from these paraphrased themes: provide the assignment rubric earlier; review the pace of worked statistical examples; retain the contrasting observation that the current pace worked for some students. Suggest changes the teaching team can evaluate and questions for follow-up. Do not invent counts, infer student characteristics, or claim the themes represent the whole cohort.
Use this workflow
- Read the feedback in the university’s authorised system and prepare a short set of themes.
- Use local detection to review any names, contacts, or student references left in the draft.
- Remove sensitive circumstances and distinctive quotations manually; preserve disagreement rather than inventing consensus.
- Ask for an improvement agenda and verify every claimed theme against the original feedback before sharing it with the teaching team.
What you should have at the end
An agenda for the course review meeting, with student-level material retained in the university’s authorised systems.
This workflow is for course improvement, not marking, admissions, misconduct decisions, or student profiling. Redaction alone does not authorise use of a new AI provider.
Questions for this workflow
Can I paste the anonymous survey export directly?
Not without reviewing it. Free-text comments can reveal names, health information, or identifiable events even when the survey does not request a name. Prepare a minimal, reviewed summary first.
Should the AI decide which student complaints are valid?
No. Use it to organise themes and possible actions, not judge students. The teaching team should examine the evidence and make decisions through its normal process.
Workflow context & further reading
These sources inform the workflow. The example and editing choices are Procurato’s illustration; they are not source-endorsed advice.
- Cornell: End-of-Semester Course Evaluations (US) — Reviewing feedback themes and deciding which course changes to make.
- Cornell: Ethical AI for Teaching and Learning (US) — Protecting student information and critically reviewing AI output.
From example to your own work
Keep the context. Review the details.
Paste into the local redactor, review each detection, choose what to keep, and check the result before sharing it with your approved AI tool. Names are only part of the review: inspect confidential wording and identifying context too.
Free, no account. Initial model download: about 1.1 GB. Requires WebGPU and several GB of free memory. Detection focuses on English and can miss details. The examples here include manual editing; the tool does not automatically rewrite or summarise your text.