Turnitin's Detection Changes Over Time: What That Means for Your Score

The same paper can receive different AI scores on different days. Turnitin's FAQ states: "As we iterate and develop our model further to better detect newer LLMs, it is likely that our detection capabilities will also change, affecting the AI percentage." The FAQ continues: "However, for a submitted document, the AI percentage will change only if it's re-submitted again to be processed." This means that once a report is generated, it stays as it is. But if you resubmit the same text later, the new report reflects the updated model's capabilities. A paper that scored 20% in January might score differently in August because the model has changed, not because the text has changed.

The FAQ is explicit about this:

"As we iterate and develop our model further to better detect newer LLMs, it is likely that our detection capabilities will also change, affecting the AI percentage.. However, for a submitted document, the AI percentage will change only if it's re-submitted again to be processed."

Two things to notice. First, Turnitin acknowledges that its detection capabilities are not static. The model evolves to detect newer language models, and that evolution affects the AI percentage the model assigns. Second, the change only applies on resubmission. Existing reports are not reprocessed. If you have a report from three months ago, that report's score is frozen. It reflects the model as it was at the time of processing, not as it is now. Which is why still flagged after a Turnitin recheck argues for comparing two reports rather than two numbers.

Three release notes from the AI writing detection model page, dated October 2025, February 2026, and May 2026, all contain the same sentence:

"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."

This is consistent with the FAQ statement. Each model update improves detection but does not go back and recalculate old reports. The direction of these updates is also stated: "We have updated our AI writing detection model to improve recall while maintaining a low false positive rate." Improving recall means the model becomes more sensitive to AI-generated text. A paper that previously scored low might score higher after an update because the updated model catches patterns the old model missed. That is one of the explanations behind editing a paper and watching the AI score go up.

The FAQ describes a significant recent change: "In July 2026, we updated our model architecture to consolidate a multi-model ensemble into a single model. This update improves and simplifies the AI writing report, maintaining a less than 1% false positive rate."

This is from the FAQ, not from the release notes page. The FAQ says a multi-model ensemble was consolidated into a single model. This is the kind of architectural change that can produce noticeably different scores on the same text. A multi-model ensemble pools predictions from multiple models, while a single model makes its own predictions. The consolidation means the scoring path changed, and with it, the classification of borderline segments may have shifted.

The model's capabilities have also expanded over time in specific directions. In December 2023, Turnitin added detection for AI word spinners. In August 2025, the detector was updated to include text modified by AI bypasser tools within the "AI-generated only" category. These additions mean that text processed through paraphraser or bypasser tools is more likely to be detected in reports generated after these updates than before. Does Turnitin detect AI humanizers goes through what the vendor will and will not commit to there.

To summarize what we have covered:

• Turnitin's FAQ says detection capabilities "will also change" as the model evolves to detect newer LLMs.
• Release notes confirm that updates do not retroactively change previously-generated reports.
• The direction of recent updates is to improve recall while maintaining a low false positive rate.
• A July 2026 architecture change consolidated a multi-model ensemble into a single model.
• The same text can produce different AI scores if resubmitted after a model update.

If you have a Turnitin report and want to work on the flagged passages, import the report and address them directly. Eligible passages can be re-run at no charge.