5 Simple Ways to Identify Text Written by ChatGPT and Other AI Tools

5 Simple Ways to Identify Text Written by ChatGPT and Other AI Tools in 2023

Introduction

Embarking on a quest to distinguish between human-crafted prose and ChatGPT’s uncanny creations has become an intricate battle of wits. The task of discerning the concoctions of human intellect from those of ingenious algorithms might be best entrusted to the digital minds themselves. Propelled by the advent of ChatGPT and its formidable linguistic brethren, the realm of detection tools has burgeoned. These tools, while mostly available free of cost, do come with certain character limitations (which are easily circumvented by dividing the text into small chunks for analysis).

Harnessing the power of an AI detector is a multifaceted endeavor. Not only does it aid in mitigating the risks of contrived and lackluster content, but it also enables unmasking potential deceit perpetrated by job applicants. Among those who would benefit significantly from a dependable method of unmistakably identifying AI-generated text are educators, who have aptly embraced such detector software from its inception. However, much like the capricious nature of ChatGPT and its kin, the reliability of AI detectors remains an ongoing challenge.

A recent incident on the ChatGPT subreddit illustrates this dilemma. A diligent high school student found themselves in a grievous predicament when falsely accused by their history teacher of utilizing ChatGPT’s assistance. Regrettably, the teacher refused to disclose the exact tool employed and, as per the student’s account, deemed the accusation justified based on the detector’s efficacy in detecting other AI-written texts, confessed to by fellow students mired in ChatGPT’s enticing world.

Before we delve into the world of AI detectors and our personal encounters with them, let us share a cautionary tale. The remarkable capabilities of ChatGPT and similar models to replicate human speech create a challenge in distinguishing between borrowed phrases from common usage and genuinely human-written text. Even AI finds it arduous to distinguish between the two. Within the AI community, there is some discussion about incorporating watermarks or detectable signals into AI-generated content without compromising its readability.

However, for this to materialize, cooperation from the AI content-producing companies is indispensable. Yet, it is highly unlikely for these companies to willingly adopt such measures that may place them at a disadvantage compared to their rivals. Nevertheless, we present to you a collection of the most prevalent AI detectors. To assess their effectiveness, I tested all the available free ones using excerpts from my own article titled “Is Dall-E the Next Dior? How AI Is Trying to ‘Make It Work’ in Fashion,” as well as text generated by ChatGPT utilizing the prompt “Please write me an article on how AI is being used in the fashion industry, specifically Stable Diffusion, DALL-E 2, and Midjourney.” In this professional discourse, we aim to shed light on the matter at hand.

5 Simple Ways to Identify Text Written by ChatGPT and Other AI Tools
5 Simple Ways to Identify Text Written by ChatGPT and Other AI Tools

1. AI Text Classifier

ChatGPT evaluating itself may appear unusual, but as an AI, it probably couldn’t care less. OpenAI, being transparent, acknowledges the limitations of this solution. It performs optimally with at least 1,000 characters, has the potential to mislabel both AI-generated and human-written text, isn’t proficient with non-English or child-authored content, and small adjustments to AI-generated text can conveniently outsmart the classifier.

The AI Text Classifier, available for free, assigns grades to text based on its likelihood of being AI-generated: either very unlikely, unlikely, or unclear. OpenAI emphasizes that the purpose of this detector is to stimulate discussions on distinguishing between human-written and AI-generated content, rather than providing absolute answers. In my evaluation, I utilized the AI Text Classifier to assess the entire ChatGPT-written essay, resulting in the classifier deeming it “likely AI-generated.” As for my own text, the tool determined it to be “very unlikely AI-generated.”

AI Text Classifier
AI Text Classifier

2. GPTZero

GPTZero had swiftly shattered the aspirations of college students shortly after ChatGPT sparked a media frenzy. This groundbreaking creation was crafted by none other than Edward Tsai, a talented Princeton senior who combined his computer science expertise with his passion for journalism. Tsai cleverly employed his knowledge to analyze the perplexity and burstiness of texts, assessing their complexity and variation in sentence length.

Through rigorous training on a multitude of human-written and AI-generated texts, GPTZero emerged as a remarkable tool capable of evaluating single sentences, provided they surpass the 250-character limit. With each text it consumes, GPTZero’s accuracy steadily ascends, delighting and captivating educators across the board.

Nevertheless, the program’s FAQ issues a necessary cautionary note, warning against the punitive use of its results on students. Educators are urged to view GPTZero’s findings as merely one component of a comprehensive assessment, mindful that rare cases may arise where AI is misidentified as human, and vice versa.

Introducing GPTZero Educator, an exclusive product tailored specifically for educators. For a monthly fee of just $9.99, educators can access a range of features designed to meet their unique needs. This includes the ability to effortlessly identify AI-generated sentences, handle larger volumes of text, and support multiple file formats. With GPTZero Educator, users can confidently check up to 1 million words every month, ensuring the utmost accuracy and efficiency.

But the benefits don’t stop there. GPTZero also offers a free version accessible to anyone at GPTZero.me. By simply registering, users can examine up to 5,000 characters per document and upload three files in a single batch. For those seeking even more advanced capabilities, the Pro plan, available for just $19.99 per month, incorporates an enhanced AI model that effectively analyzes a higher volume of words each month.

Curious about the accuracy of AI-generated text? GPTZero has got you covered. When I fed the AI-written text into GPTZero, the results confidently stated, “Your text is likely to be written entirely by AI.” However, my own content received a contrasting outcome, with GPTZero affirming, “Your text is likely to be written entirely by a human.” Furthermore, GPTZero effectively distinguished the ChatGPT-generated text as AI-written, providing it with an average perplexity score of 31.929 and a burstiness score of 21.385. In contrast, my article received no indications of AI-generated sentences, scoring an average perplexity of 588.703 and a burstiness score of 2,545.787. While the precise methodology behind GPTZero’s assessment of burstiness and perplexity remains unknown, the results undeniably reflect a positive outcome.

GPTZero results
GPTZero results

3. Originality.AI

Introducing Originality.AI, an innovative solution tailored for the business world, not academia. This cutting-edge product is designed to assist content publishers in safeguarding their integrity by ensuring that source material or submitted works are free from plagiarism or AI-generated content. With the power of its Chrome extension, Originality.AI swiftly scrutinizes emails, Google docs, websites, and more.

Setting itself apart from other AI detectors, Originality.AI proudly boasts an exceptional detection rate. With an impressive 96% overall detection rate, it confidently achieves a remarkable 99%+ accuracy on GPT-4 content, 94.5% on paraphrased content, and 83% on ChatGPT. How does it achieve this unparalleled performance? By harnessing immense compute power and foregoing a free option.

When it comes to pricing, Originality.AI adopts a simple and automated approach. For a mere one cent per 100 words, users can effortlessly utilize its services, with automatic billing taking care of the rest. It is important to note that, since Originality.AI does not offer a free tool, we regretfully did not have the opportunity to put it to the test with our own content.

4. Writer AI Content Detector

Inspired by the AI-writing phenomenon, the ingenious Writer AI Content Detector has emerged, offering a streamlined approach to identifying the authenticity of written content. Devoid of complexity yet brilliantly effective, all you need to do is input a URL or up to 1,500 characters into the designated box and witness a substantial detection score in all its magnificence. The best part? This ingenious tool is completely free of charge. However, for those privileged souls with a Writer enterprise plan, the opportunity to engage in discussions regarding large-scale detection awaits.

Eager to put this remarkable creation to the test, the detector examined a 1,500-word piece written by the ChatGPT AI, promptly labeling it as “0% human-generated content.” Unfazed by this seemingly harsh judgment, it kindly advised the user to polish the text until the discernible AI presence was minimized. On the other hand, a piece I composed myself faced a much more favorable fate. Acquiring a “100% human-generated” score, the detector’s robotic counterpart showered it with praise, exclaiming a resounding “Fantastic!”.

Writer AI Content Detector results
Writer AI Content Detector results

5. ZeroGPT

Introducing ZeroGPT, a cutting-edge tool that caters to a wide range of users, from students and teachers to writers and freelancers. With an impressive accuracy rate of 98%, ZeroGPT utilizes its proprietary technology, known as DeepAnalyse, to provide users with insightful analysis. This technology has been honed through extensive training on a vast dataset of 10 million articles and text.

Using ZeroGPT is a breeze. Simply paste your text into the designated box on the website, and let the magic unfold. Prepare to be amazed as ZeroGPT presents you with one of several intriguing results: Is the text human-written or AI/GPT-generated? Perhaps it falls into the realm of mostly AI/GPT-generated or most likely AI/GPT-generated. Alternatively, it could display mixed signals, blending human and AI elements.

ZeroGPT possesses an uncanny ability to discern AI influence in your text. It astutely identified my submission as AI/GPT-generated, with an impressive 94.03% accuracy. Remarkably, it even pinpointed which sentences were written by AI, save for one. On the other hand, I felt a wave of relief when ZeroGPT confidently asserted that my writing was indeed human-written, registering a perfect 0% AI-written score.

ZeroGPT results
ZeroGPT results

Leave a Reply