Flag suspiciously similar submissions before you finish grading a question.
From inside ZenGrader, click Check Similarity. This compares every enrolled student's submission for the current question against every other student's, looking for suspicious textual/structural overlap.
The report lists pairs of students whose submissions crossed a similarity floor, each with:
| Column | Meaning |
|---|---|
| Jaccard % | Overlap based on shared words/tokens between the two submissions. |
| Cosine % | Similarity of meaning. Answers to the same question can score highly without copying; this is not a probability of misconduct. |
| Review priority | High / Medium priority reflects substantial wording overlap or a Likely Copied / Possible AI verdict. An Independent verdict gives Low priority. Other pairs without an AI verdict are Unassessed; semantic similarity alone does not raise their priority. |
| AI verdict | For the closest pairs, an AI-generated read on whether the overlap looks like genuine copying versus coincidental similarity (e.g. both citing the same textbook definition), with a confidence level. |
| Status | Your review status for this pair: pending, reviewed, dismissed, or flagged. |
Each completed report shows Pairs needing review (High/Medium priority pairs you haven't reviewed), Pairs with review signals (all High/Medium priority pairs, including those already reviewed), and Flagged to student (deliberate, student-visible flags you've raised). Independent verdicts are excluded from the first two counts.
Pair counts are neither student counts nor percentages: one student can appear in several pairs. A report retains up to 100 candidate pairs and states how many have an AI verdict. For example, 29 submissions from different students produce 406 possible comparisons, but at most 100 pairs appear in the report.
Click Details to expand a pair's evidence, a shared text snippet and/or the AI's rationale for its verdict, when available. Mark the pair Reviewed, Dismissed, or Flagged for follow-up from the same row.
Click Compare on any pair to open both submissions together in their own window. It has two tabs.
| Tab | What it shows |
|---|---|
| Documents | Both PDFs rendered side by side, with zoom and an optional Sync scroll that moves the two panes together. |
| Matching text | The text of both submissions, with every passage that appears in both highlighted. Each pane's heading gives the share of that submission's words that are shared. |
Highlighting marks runs of eight words or more that both submissions contain. A shared cover page, a restated question paper, or supplied code highlights the same way copied work does, which is usually the fastest way to see that a pair matches only on material you handed out. Scores, the review priority, the AI verdict and the Reviewed, Flag and Dismiss buttons are all in the window's header, so you can triage without going back to the report.
Each student's heading also links to Open for the original file and Grade to take that submission into ZenGrader.
Every course has an Integrity Center (open it from the AI Similarity & Integrity card on your course dashboard). It's the hub for this feature: run a check for any module or question that has two or more submitted files, see each assessment's latest status and count of pairs with review signals at a glance, and open the full history of every report you've run, all in one place. You can still launch a check from inside ZenGrader; both land in the same reports.
Reviewing a report is private to you. If you decide a submission warrants it, expand the pair and choose Flag Student to raise a student-visible flag. The student then sees a calm, non-accusatory notice on their course page asking them to review and follow up, with your optional message. You can also tick send a follow-up message, which delivers it through the normal course messaging (and emails them if they don't check the dashboard). Raising a flag never changes a grade, and you can resolve or withdraw it at any time.
Like all AI features in ProfDesk, similarity detection is a flag for your attention, not an automatic accusation. A high score means "look closer," not "guilty." Shared boilerplate, common formulas, or standard code patterns can all trigger overlap without any misconduct. You decide, case by case, what a flagged pair means.
