The emergence of various models of artificial intelligence detectors was not long in coming after the emergence of more advanced generative AI.
With the rise of tools like ChatGPT or Google Play, tools that try to detect when a text could have been generated by artificial intelligence are useful for many people: from teachers to moderators of websites or content publishing platforms.
As we’ll see below, questions have also arisen about how to use an AI-generated text detector or whether these tools are truly useful or reliable.
In this regard, most expert analyses have established that, as of the end of 2024, these types of tools are still experimental platforms and therefore not 100% reliable. For example, an analysis by Scribble found that the best paid platform had an 84% accuracy rate, while the best free tool had an accuracy rate of 68%.
Precisely for this reason, below we provide an analysis of how an AI text detector works and some considerations regarding its use.
What is an AI content detector?
An AI content detector is a tool designed to identify whether a text has been generated by generative AI tools, such as ChatGPT.
How does an AI content detector work?
The basic operating premise of an AI content detector is as follows: the AI tester analyzes a text and its characteristics ; its goal is to see if it resembles the human text samples it was trained on or if, on the contrary, the text is more similar to the AI-generated texts it was trained on.
To achieve this, the various AI detection models have often been trained with language models similar to the AI tools that generate text.
The following two features are the two main markers that AI detectors typically use to determine the likelihood that text is human-generated:
- Perplexity: This measures how unpredictable a text’s content is. AI-generated texts are generally considered to have low perplexity (they are more predictable); human-generated texts have high perplexity (due to the use of figures of speech, possible typographical errors, possible gaps in the subject matter, etc.).
- Explosiveness: This focuses on variations in sentence structure, in addition to variations in length. AI-generated texts are generally considered to have low explosiveness (little variation in sentence structure and length) compared to human texts.
Important Note: Easily submit your business listing online and reach more customers. Boost visibility, attract leads, and grow your brand with our trusted platform.
Why it’s important to use an AI content detector
Due to their relative novelty and lack of policies for their use, AI text generation tools have proven disruptive for some sectors.
On the one hand, copyright concerns are at the heart of the controversies surrounding these types of tools and, therefore, also among the reasons for using an AI detector. This concern has already been addressed by the EU Artificial Intelligence Act , which, among other measures, establishes that these types of tools must review their models to avoid violating intellectual property laws.
On the other hand, educational institutions have also sought to turn to AI detection tools to prevent potential fraud or plagiarism.
Likewise, the emergence of these tools in the field of content creation and digital marketing has raised several issues. Among them, the fear of possible Google penalties on the SEO ranking of content generated by AI stands out.
In this regard, Google responded to these concerns as early as 2023: in an article titled “We reward high-quality content, no matter how it’s produced,” the search engine explained that its algorithms would continue to “prioritize the creation of valuable content” regardless of how it was created. In the article, Google asserted that AI tools can serve “as a critical tool to help people create quality content for the web,” as long as they are not used for fraudulent practices, such as spam.
When to use an AI content detector?
Using an AI content detector can be useful to provide clues about the likelihood that a text was generated by AI, with the nuances outlined below.
Overall, at the time of writing this article, the following AI content detection models stood out among the lists analyzing the effectiveness of these tools:
- Copyleaks
- ZeroGPT
- GLRT
- Undetectable.ai
- AI Originality
- Compilatio
- Winston AI
- Content at scale
- DetectGPT
The use of artificial intelligence detectors must be qualified, however, by the results these tools are capable of achieving. In this regard, it is important to take into account the inaccuracies found in the results of AI text detector models.
In other words, as of today, no AI tester can be used as proof that AI was used to create a text.
Among the best practices for using an AI detector, it’s helpful to learn about publicly available effectiveness analyses of these tools. You can also conduct personal tests to see firsthand whether the results are satisfactory.
Do you want to learn more about responsible uses of AI and how to harness the full potential of these technologies ethically and legally? At City Local 101, we help SMEs and freelancers incorporate artificial intelligence tools that will help boost their businesses. Discover our solutions!