Turnitin uses its own proprietary AI Writing Detection tool to identify AI- generated text in student submissions. This AI detector works by analyzing segments of text with a proprietary algorithm trained specifically on AI models like GPT-3.5, which is a version of ChatGPT. The tool assigns a score indicating the likelihood that a portion of the text was generated by AI, with a confidence level of about 98% when it flags AI writing. The detection results appear as an additional indicator within Turnitin's Similarity Report interface, showing a percentage or an asterisk to denote the estimated amount of AI-generated content
. Moreover, Turnitin's AI detection model is based on an open-source foundation model developed by Huggingface, which Turnitin has trained, evaluated, and fine-tuned multiple times to improve accuracy
. Despite its sophistication, the tool is not perfect and can produce false positives or miss some AI-generated text. Turnitin advises educators to use the AI writing indicator as one piece of evidence rather than a definitive judgment
. In summary, Turnitin uses a proprietary AI detection algorithm trained on models like GPT-3.5, built upon an open-source foundation from Huggingface, to detect AI-generated writing in student papers.
