Reply to the Comments on “How Sensitive Are the Free AI-detector Tools in Detecting AI-generated Texts? A Comparison of Popular AI-detector Tools”

Dear Commentators,

Artificial intelligence (AI) has been increasingly used in academics and research in recent years. Generative AIs instantly produce text contents and images and translate information into academic research. The use of generative AI has become so common that students, researchers, and academicians are inclined to use it to meet their academic and research needs.^1,2 Generative AI tools produce lucrative content in reasonably less time. Complete reliance on generative AI tools for producing academic content without exercising one’s skills and knowledge is considered scientific misconduct. ³ To combat the issue of generative AI-related misconduct, several AI content detection tools have been developed in recent years. In our recently published study, we attempted to identify the free AI detection tools that may be of value in detecting AI-generated texts. ⁴ We identified the first 10 tools that appeared on the Google database (the appearance of these tools is mostly on the basis of their popularity in that database). Several readers have shown interest in this research.

Meanwhile, some queries related to the article were raised. It was pointed out that the study needs selection criteria for the software evaluated in this research. Also, it was highlighted that the different tools may have distinct AI content detection potential, and a single prompt may not adequately evaluate the capabilities of the tools.

There are multiple factors to consider while defining the criteria for selecting AI content detection tools. The selection can be based on their popularity, the reviews and ratings of the users, the order in which they appear on the search platforms, and the pricing and availability of the tools. However, all the mentioned factors are amenable to rapid change owing to the frequent upgrades in the software and the terms and conditions of their use, the rapid evolution of the software market, and the launch of new software for the function. Ours was preliminary research highlighting the difference in the sensitivity of different freely available tools at a given time point. The primary message conveys that the text contents should be analyzed using various tools available in their free versions prior to making any determinations regarding whether the text constitutes AI-generated content. The cost considerations preclude using paid versions of the tools for comparison in the research. However, further research addressing these limitations was endorsed. We concur with the comment that testing diverse text types, such as narratives and conversations, generated through AI may impart more credence to the findings. Future research concerning AI detection tools should evaluate different text types. Again, there is a risk of false-positive (human-generated text labeled as AI-generated) and false-negative (AI-generated text being labeled as human-generated) detection of AI-generated text, which needs to be explored in future research. To complicate the issue of AI-generated content, there is a new group of software that humanizes the content by making it look like humans generated it. Maybe in the days to come, there will be development software to detect AI-assisted humanized text content.