Why Did My AI Score Change When I Resubmitted the Same Paper?

When the same paper comes back with a different AI percentage the second time you submit it, the most common cause is not that your text changed. It is that the detector did. Turnitin updates its AI writing detection model on a rolling basis, and each release note carries the same sentence: "This release will not retroactively update previously-generated AI writing reports." A report is a reading taken by whatever model was in force on the day it was generated. If a model update falls between your first and second submission, the two scores were produced by different versions of the detector, and nothing in either document tells you how much of the gap is your edits and how much is the model.

People tend to assume that "Turnitin" is one fixed instrument that always reads the same way. The release notes tell a different story. The AI writing detection model page records updates on 14 October 2025, 12 February 2026, and 5 May 2026. Each of those entries says:

"This release will not retroactively update previously-generated AI writing reports. To view the AI writing score generated by this update, existing submissions must be resubmitted."

The February and October notes also share a direction: "We have updated our AI writing detection model to improve recall while maintaining a low false positive rate." The FAQ adds that as the model iterates, "it is likely that our detection capabilities will also change, affecting the AI percentage." Turnitin says this plainly. The detector is a moving target, and a score pinned to a date is not a property of the document. It is a property of the document as read by that version of the model on that day.

The FAQ also records a more recent architectural change: "In July 2026, we updated our model architecture to consolidate a multi-model ensemble into a single model." The FAQ says this update maintains a less than 1% false positive rate. We note this because a student who submitted in June and resubmitted in August may have crossed a model boundary that is larger than the usual incremental update.

Say your first report came back at 28 percent, you revised a few sentences, and the recheck says 42. The instinct is to ask what went wrong with the revision. The first question to ask instead is whether a release date falls between the two runs.

If one does, the gap contains your edits and a model change mixed together, with nothing in either report that separates them. Fourteen points is not fourteen points of revision. It is fourteen points of revision plus a model, and the model half is not yours to claim or to blame. Comparing the two reports rather than the two numbers is the whole method in still flagged after a Turnitin recheck. The honest comparison requires writing down three things: the date on report one, the date on report two, and which release notes fall in between. Turnitin publishes all of them on its AI writing detection model page.

The release notes we mentioned share a phrase: "improve recall while maintaining a low false positive rate." Recall is the proportion of AI-written text the detector catches. Improving recall means the model is being tuned to catch more, not less. The FAQ explains the trade-off this creates: "In order to maintain this low rate of 1% for false positives, there is a chance that we might miss some AI written text in a document." And: "if we identify that 50% of a document is likely written by an AI tool, it could contain as much as 65% AI writing."

The direction matters. A model update that improves recall can raise the AI percentage on a paper that did not change at all. That is not a bug. It is the model doing what it was updated to do. Conversely, a rewrite that lowered the score might have been helped by a model update that happened to read your text differently, not by the edits themselves. Without the dates, you cannot tell which is which. The version of this most people run into is editing a paper and watching the AI score go up.

One more thing that surprises people: the AI percentage and the similarity percentage are separate. Turnitin states: "The Similarity score and the AI writing detection percentage are completely independent and do not influence each other." A change in one does not predict a change in the other. If your similarity score dropped but your AI score rose, that is not a contradiction. It is two systems reading different signals in the same text. We have written about this distinction in more detail in our guide to AI Writing Report vs Similarity Report.

To summarize what we have covered:

• Old reports are not re-scored when the model updates. A score is a snapshot, not a permanent property.
• If a model update falls between two submissions, the two scores were produced by different detector versions.
• Recent updates have improved recall, meaning the same text may receive a higher score under a newer model.
• The AI percentage and the similarity percentage are independent and do not move together.

The practical step is to record the date on each report and check it against the release notes before comparing two numbers. If you need to revise a flagged paper, import your Turnitin report and work on the specific passages that were flagged. Eligible passages can be re-run at no charge.