Turnitin’s new similarity report feels different this year, anyone else notice?
ran my first batch of turnitin reports for the year yesterday and something feels off. scores are lower across the board compared to june, like noticeably lower. either the algorithm changed over the summer or my students got way better at hiding stuff over the break lol. anyone else running reports yet? i teach grade 11 english and usually i get a handful of red flags in the first week just from kids reusing old essays from other courses. this year almost nothing. not sure if thats a good sign or if the tool just got worse at catching things. curious if other people noticed changes to the interface too, mine looks slightly different, new colour coding on the percentage bar. wondering if this is a stealth update they didnt announce. would love to hear from anyone in a district that got a memo about changes this summer.
5 Replies
Join the discussion.
Log In to Reply
worth being cautious here. as a vp i've seen what happens when staff assume a lower score means less oversight needed. we had a parent threaten legal action last spring after a teacher used a single turnitin percentage as the sole basis for an academic integrity meeting. the tool changing its baseline doesnt change your obligation to actually read the writing and use professional judgment.
This matches what I have read about the recent update. Turnitin quietly recalibrated its algorithm in June to better distinguish between properly attributed paraphrase and genuine overlap. The practical implication for classroom teachers is that the percentage itself was never a reliable single indicator of academic dishonesty, and that remains true now. I would treat the lower numbers as a return to a more accurate baseline rather than evidence that detection has weakened. Pair the report with a conversation with the student before drawing conclusions.
@Patrick Lefebvre that actually makes me feel better honestly. i was half convinced i was gonna have a repeat of last year where three separate parents called the office in the same week. gonna keep doing the conversation first approach either way.
tested walter writes ai on a few of the flagged samples out of curiosity. it flattened out most of the paraphrase heavy ones as human, which lines up with what patrick's saying about the recalibration. still wouldnt use any single tool as a verdict though.
YES!! i noticed this too. our board sent out a small update notice in july, said they retrained the similarity model to reduce false positives on properly cited paraphrasing. so lower scores might actually mean its working better, not worse. i was thrilled because last year i had at least three kids in tears over flags that turned out to be legit citations.