Turnitin Adds AI Writing Indicator to Similarity Report 

Reading Time: 3 minutes

It’s no secret that ChatGPT’s popularity is growing. And plagiarism detecting tools are aware of that – and the need to adjust their offerings to help identify when student work may have been supplemented or entirely created by AI. Turnitin recently announced that it will now display a score to instructors within the Similarity Report to indicate what percentage of a paper is highly likely to be generated by AI. 

Turnitin is quick to say that their detection isn’t foolproof and the percentage displayed “should not be used as a definitive measure of misconduct or punitive tool.” The AI percentage is presented to inform instructors and give them data so they can decide how to then handle the situation. It should also be noted that this percentage is only visible to instructors – students will not be able to check their own papers for this AI score.  

How does Turnitin’s AI detection work?  

Turnitin can only detect AI writing in long-form paragraphed writing. It’s not able to reliably identify AI influence in short-form content (bulleted lists, tables, annotated bibliographies) or non-prose writing (poetry, scripts, code). Turnitin examines each sentence in a submitted work in context – it chunks the work into about 100-word segments, then overlaps them to ensure each sentence is checked in its proper context. Each sentence is then internally labeled as AI or not-AI, and that is used to create the AI-generated percentage. 

How can I find this AI score?  

The AI percentage is located below the standard similarity score in the Similarity Report. The Similarity Report, however, does not show the highlighting for what sections Turnitin has identified as likely generated by AI. That can be accessed by clicking on the AI percentage and opening the AI writing report.

What can I learn from the AI Writing Report? 

The AI writing report shows the percentage of AI-created material within the paper and highlights those sections. I asked ChatGPT to write a blog post on creating clear content, and Turnitin correctly identified it as AI-generated.  

I also ran a blog post on the same topic that we’ve previously published on FOCUS, and Turnitin again correctly identified it as human-written (though the similarity report was very concerned that this exactly matched the FOCUS blog post). 

And finally, I was curious about how Turnitin would handle a paper that started out as generated by AI and was then edited or partially rewritten. I took the ChatGPT-generated blog post on creating clear content and rewrote large sections of it, primarily rephrasing and simplifying things. If I left the numbered lists from the beginning in the post, Turnitin detected less AI influence (the AI detector focuses on prose – it doesn’t work well with lists, scripts, poetry, etc). But once I reworked that list to be headings and paragraphs, Turnitin detected that 35% of the paper was generated by AI.  

There’s so much more that could be said about this new feature from Turnitin. To read their full notes and FAQ, check out Turnitin’s AI Writing Detection Capabilities page.  

Want to never miss a post? Subscribe here!