Start with authorship decisions, not suspicion
Students need more than a definition they can repeat in a quiz. They need to recognise when a claim depends on a source, represent that source accurately, distinguish borrowed language from their own wording, and explain what they add. Those judgements vary by discipline. Source practice in history, a laboratory report and a legal problem answer will not look identical.
Begin with the real seminar assignment. Identify learning outcomes, permitted collaboration, expected source types and the required citation system. Explain which tools may be used and how relevant assistance should be acknowledged. Only the rules genuinely adopted for that course and institution belong here. A teaching handout must not create prohibitions or penalties that the approved assessment documentation does not contain.
Use a low-stakes diagnosis and complete examples
Before week one, ask students to inspect two short cases: an overly close paraphrase and a factual claim without a clear source. Use the responses to plan teaching, not to punish. They may reveal uncertainty about attribution, reading comprehension, time management or disciplinary conventions. Do not turn one diagnostic response into a judgement about a student’s character or future conduct.
Examples should be synthetic or licensed for this use and include complete bibliographic information. A sentence pair without context can train superficial pattern matching. Supply the source passage, student passage, assignment prompt and expected citation style. Provide accessible formats and enough time for students who use assistive technology or work in an additional language.
A workable starting allocation is 25–40 minutes of class time plus a small preparation task each week. Adjust this to the course credit and timetable rather than presenting it as a universal dose. Feedback may come from an instructor, structured peer exchange and annotated model answers. Incorrect answers should not remain uncorrected, and private drafts should never be displayed to the class without informed permission.
Attach feedback to observable choices
“Add more citations” is rarely enough. Point to a location, name the expected action and explain why: “This sentence adopts the source’s specific classification; attribute it here, then explain why the category applies to your question.” That response connects text, evidence and learning outcome. A similarity percentage cannot make that judgement for the instructor.
Peer review needs one narrow task. A peer might check whether a paragraph lets the reader distinguish sourced material from the writer’s inference. They do not infer intention or assign a misconduct label. Students should be able to use a supplied practice passage instead of sharing sensitive work. The instructor moderates disputed examples and corrects false rules promptly.
A simple feedback grid can distinguish “clearly supported,” “source present but relationship unclear,” “paraphrase too close,” and “claim lacks identifiable support.” It remains a learning device, not an automatic grade. Different wording may satisfy the same criterion. If any artefact contributes to assessment, its criteria and weighting need to be transparent in advance.
Move from exercises into the student’s own project
In week four, each student selects a manageable section of a current draft. A compact source matrix records the claim, location in the source, type of use and the writer’s own inference. The student then compares an earlier and revised passage. A short note identifies two deliberate changes and one unresolved question; it does not demand continuous surveillance of the writing process.
Students without a draft receive an equivalent prepared passage. Those using assistive technology, translation or permitted AI support follow the course’s announced disclosure rule. Tool use alone is neither misconduct nor evidence of quality. Requiring one commercial platform can also create an access barrier, so equivalent local or non-digital methods should be available where appropriate.
Close with an anonymised discussion of common choices. A real allegation concerning an individual does not belong in a classroom activity. If a separate concern arises, route it through the institution’s fair process. Keeping teaching and investigation distinct makes it safer for students to reveal genuine uncertainty during practice.
Measure the specific learning, not a sweeping promise
Before delivery, choose two or three indicators: can students explain why a claim needs evidence, repair an overly close paraphrase, and describe what a cited source contributes? Use comparable short tasks at the beginning and end. Improvement supports a claim about performance on those activities, not a general claim that long-term plagiarism has fallen.
Add anonymous process feedback: which activity informed a real writing decision, which instruction remained ambiguous and what access problem arose? Record the resulting changes for the next cohort. Do not publish a success percentage without cohort, task, scoring and missing-response context. Where sample sizes are small, qualitative patterns may be more honest than a headline number.
The citation-diagnosis case set supplies ready-to-use material for week three. The independent-work rubric helps separate process and product evidence. Find more tools in the higher-education practice hub, and use the evidence-review foundation for careful interpretation of reports.