A repeatable review workflow
Step one: collect the submission and run analysis with submitted-content-only PDF mode when Turnitin wrappers or cover pages are present. Step two: read the overall AI band and confidence label. Step three: inspect highlighted passages — not every orange segment needs a meeting.
Step four: compare against draft history when available. Step five: record outcome as cleared, revision required, or referred to formal integrity process. Keep notes factual: which passages, which questions asked, what evidence the student provided.
Rubric language that fits probabilistic tools
Avoid rubric rows that say 'must score 0% on AI detector.' Instead: 'Writing shows original analysis tied to course sources; generic or unattributed AI-like passages are revised.' This aligns assessment with learning while acknowledging detector limits.
Separate content quality from integrity suspicion. A weak essay is not necessarily an integrity case. A polished generic essay may warrant questions even when scores are moderate.
Batch review for large courses
Team-plan batch upload lets you queue classroom sets with per-file summaries. Sort by band to prioritize review time, but do not auto-penalize top buckets. Use batch export for department meetings where you need aggregate patterns, not individual verdicts.
Pair batch scans with a single clarification email template so students know elevated scores trigger conversation, not automatic failure.
Templates you can adapt
Office-hour invitation: 'Your submission flagged a few passages for style review. Please bring prior drafts or notes so we can discuss how you developed the argument.' Appeal response: 'We considered your draft history, the flagged excerpts, and your explanation of sources.'
Department memo: 'Automated scores are one input. No grade penalty or integrity finding may rest on a score alone.'
Pilot and professional development
Explanation Panel and AI Paraphrase Risk are Pro plan features. Join the pilot if you want early access to upcoming features and to help calibrate detection with classroom feedback.
Schedule a short department session on reading PDF reports: cyan highlights for AI-generated signals, purple for paraphrase-risk when confidently detected, and gray zones when attribution fails.