Skip to main content

AI writing detection model

On this page, review any information or notes related to our most-recent releases for Turnitin's AI writing detection model.

2024 release notes
2023 release notes

2024 release notes

September 2024

2024 September 12

New AI detection writing detection capabilities for Spanish submissions

Turnitin is excited to release its AI writing detection capabilities in Spanish to help educators uphold academic integrity for Spanish language submissions. 

With this release, our AI writing detector can process submissions in Spanish for text that was likely AI-generated. Once submitted, a Spanish AI writing report is generated for instructors to review. Previously submitted Spanish assignments can be checked with our AI writing detection when they are re-submitted to Turnitin and if you have AI writing enabled for your account. 

Please note our Spanish AI writing model is a different model than our English AI writing model. Turnitin’s Spanish AI model has detection capabilities that have been trained with Spanish language writing and to detect text likely generated with the GPT-3.5 and GPT-4 large language models. As our Spanish and English AI detection capabilities differ, please read more about how our Spanish AI detection capabilities work


July 2024

2024 July 16th

AI writing detection updates

Interactive detection categories

With this release, the AI writing report now displays a submission’s overall percentage of text detected as AI and a breakdown of that score in two interactive categories:

  • AI-generated only - qualifying text that was likely generated from a Large Language Model (LLM).
  • AI-generated text that was AI-paraphrased - qualifying text that was likely AI-generated and then was likely revised using an AI-paraphrasing tool or AI word spinner, such as Quillbot.

These categories provide more granularity into what AI capabilities were used and where, as well as their frequency within the submission. Selecting a category will bring the first highlight detected in the submission into focus.

Increased maximum word count for detection

The AI writing report now supports submissions with up to 30,000 words of qualifying text.  

Scores below 20% are no longer surfaced

False positives (incorrectly flagging human-written text as AI-generated) are a possibility in AI models. To avoid potential incidence of false positives, no score or highlights are attributed for AI detection scores in the 1% to 19% range. When AI is detected below the 20% threshold in the report, it is now indicated with an asterisk (*%) and no percentage is attributed. This update goes into effect for all reports generated after this release; reports generated previous to this release will still display scores between 0-19 with an asterisk. 

Updated AI writing detection chart in admin stats

Users will also see an updated AI detection chart in the administrators statistics page. Updates to the chart include new contextual help and a separate data column for AI reports with no percentage attributed (percentages within the 1-19% range), indicated with an asterisk (*%). 


2023 release notes

December 2023

2023 December 15th

AI writing detection processing bug fix

We have fixed a prose text, long-form writing format bug that was inadvertently introduced in the December 6th update. This bug allowed submissions that did not meet the file requirements (less than 300 words) to be processed and given an  AI  writing detection score.

This issue has now been resolved.

Some  AI  reports generated in this period (6th December to 15th December 2023) did not meet the file requirements but generated an  AI  writing detection score. Results show that our accuracy increases with a little more text so submissions that cont ai n less than 300 words may result in an  AI  writing score that is likely less accurate. Please be reminded that an  AI  Writing score should not be used as the sole basis for adverse actions ag ai nst a student.

2023 December 6th

AI writing detection updates

AI  word spinners are sometimes used to avoid identification of  AI -generated text. We now detect likely  AI -generated text even if it may have been paraphrased using an  AI  word spinner.

Our  AI  paraphrasing detection is incorporated into our  AI  writing detection capability and does not require any change to your settings. There are no visual changes to the report, or the highlighting in the report, with this update. Previous reports will need to be resubmitted to detect likely  AI  paraphrasing.

August 2023

2023 August 9th

AI writing detection updates and bug fixes

Processing text in tables

We are now able to process long-form prose text in tables. Resubmit to reprocess existing submissions that cont ai n tables.

Maximum sentence length

We have increased the maximum word count for a sentence to include lengthy sentences when processing the report.

Bibliography

We have fixed a bug that was occasionally highlighting  AI  writing within references listed in a bibliography. Bibliographies are now excluded when processing the  AI  writing report. Resubmit to reprocess existing submissions that cont ai n highlighted reference sections.

May 2023

2023 May 24th

Updates to address our customers' false positive concerns

Improvements in detection at the start and end of documents

Since launch, we have observed a higher incidence of false positive detection in the first few or last few sentences of a document. Many times these sentences consist of introduction or conclusion content written in a generic way. As a result, we have changed our detection logic to help reduce these false positives.

We also worked on making our segment boundaries detection more precise which could lead in some rare cases to change of boundaries compared with a previous version.

Fix for bugs in text formatting pipeline

We have fixed a bug that was causing inconsistent text formatting that was impacting the  AI  writing percentage. This fix ensures that the model provides more consistent and accurate determinations.

Was this article helpful?
62 out of 107 found this helpful

Articles in this section

Powered by Zendesk