How to Detect an AI-Generated Text

The rise of AI tools is another facet of the amazing times we are living in, where with a few clever prompts, the AI can generate code, images, texts, and much more. However, this rise in AI tools has exposed a complex issue, where it is becoming indistinguishable to detect if something has been done by AI or by a human. 

This is where AI detection tools come into play, but keep in mind that most of them are also commercialized, similar to AI generative tools. With these tools, people can “uncover” whether a text, code, or image has been created by AI and form opinions about the content they’re consuming. A lot of questions spring to mind when exploring this topic, such as: Why would you care? How AI generates content? How does Google rank this content? We will explore these questions and many more throughout this text, so read on!

Reading Time: 4 minutes

ai-generated text

Illustration: Lenka T

How does AI generate content? 

In a nutshell, generative AI uses machine learning processes to sift through a mind-boggling number of visual and textual data, mostly online, to figure out what things appear near each other and thus form algorithms. These algorithms distinguish things of interest that the text or image generation AI will use. Said more simply, AI creates output based on the huge amount of data it is trained on, and based on a user’s prompts, it creates output that falls within the probability spectrum as determined by the data it has been trained on.  

Keep in mind that this is a simplified explanation of how AI works, but it is relevant to our exploration of the topic.  

Why should you care? 

It seems that more and more AI-generated content is popping up on reputable sites as an alternative to “traditional” content generation (i.e., written by a content writer or a blogger). However, it is not as straightforward as you may believe. While AI can pose great questions and lead content writers down a certain exploratory path, the texts it creates can oftentimes contain factual inaccuracies, which then require human input. It seems that sites are utilizing AI-generated content to some extent, but couple it with a human writer who then refines it to make more sense and to be in line with the current trends and popular topics.  

As a consumer of online content, you should care whether what you are consuming is factually correct and whether it provides any value to you. This is the primary reason why you should care about AI-generated content since sites that use it indiscriminately come across as not having your best interest in mind and not double-checking the content that they’re putting out.  

How Google treats AI-generated content?  

Officially, Google is not banning or penalizing AI-generated content, which could be one of the reasons why we’ve seen such a proliferation of AI content. This is not by all means a bad thing, as Google pays more attention to the usefulness of the content rather than who or what has created it. One of the adages Google uses is: “Rewarding high-quality content, however it is produced.” 

In essence, this translates into Google putting human needs as their first priority, which means content that is relevant, original, of high quality, and authoritative among other users, will rank highly, regardless of who or what wrote it. For those wishing to create a website and fill it out with ChatGPT content and automate the entire process without spending the time to refine the content, Google will punish them.  

How to detect AI-generated text 

There are numerous tools online that you can use to check whether the text is plagiarised or AI watermarked (i.e., created by an AI). Of course, most of them are trying to sell you their services, but there is an increasing number of completely free or trial versions that you can use to quickly test the text created by your favorite website, content creator, or even student if you are a teacher.   

AI detection involves using machine learning and natural language processing to identify patterns and determine if the content is generated by AI or humans. AI detection tools analyze context and predict the likelihood of words based on patterns. If the text follows predictable patterns, it is likely AI-generated, whereas human-written sentences tend to be more varied and creative. 

Repetition of words and lack of depth are indicators of AI-generated text. AI-generated content may also contain inaccurate or outdated information and have a more streamlined and uniform format and structure compared to human-written text. Additionally, human-written text often includes typos, informal language, and slang. AI detection tools can also analyze images and videos using pixel anomalies to detect AI-generated content. 

Testing AI detection tools 

To test out the hypothesis that the tools online can help you identify whether a text is AI-generated, we’ve turned to ChatGPT for content creation, and below, you will find the test of online tools, which also represent recommendations for usage.  

We asked ChatGPT to write a blog post on cats using the following prompt: “Assume the role of a content creator and write a blog post about cats, focusing on their social characteristics, why humans like them, some pro’s and con’s of owning cats, how to properly care for one, and give a heartwarming conclusion. Create a 1000-word blog post on the above.” 

chatgpt

ChatGPT. Source: Author screenshot  

Content at Scale 

The first tool we used is the Content at Scale AI-detector, which showed the following results when we plugged our cat blog in

content at scale

Content at scale results. Source: Author screenshot  

Hive 

The next tool we used is the Hive Ai-detector, which showed the following results when we plugged our cat blog in. 

hive

Hive results. Source: Author screenshot  

For this blog, Hive did a much better job, showing that it is 99.9% AI-generated.  

GPT Zero

The next tool we used is the GPT Zero Ai-detector, which showed the following results when we plugged our cat blog in.  

gpt zero

GPT Zero results. Source: Author screenshot  

GPT Zero showed that the text was entirely written by a human, which means that it failed during this test.

Copyleaks

  The next tool we used is the Copyleaks Ai-detector, which showed the following results when we plugged our cat blog in.  

copyleaks

Copyleaks results. Source: Author screenshot  

Copyleaks also showed that the blog we created was a human text, also failing this test.  

Conclusion  

While AI-generated content is proliferating and is set to expand even more, it is important for us to understand how it is created and what to pay attention to when consuming such content. Google seems to have the best idea, i.e., content that is relevant will get consumed and provide value, regardless of who or what wrote it, which is what we, as users, should also be looking for.  

If you are a company or a teacher, who wishes to check whether a text is AI generated, there are online, free tools that you can use, but as you have seen from our test, they’re not perfect. Some will give you mixed results, and others will show high levels of accuracy, but in the end, it comes down to the professionalism, trust, and quality of content that you are looking for. If your writers use proofreading and grammar tools, the AI alarm may be triggered if some of the sentence structure is changed by such tools.   

Finally, Content at Scale and Hive seems to be the best option if you are overly worried about consuming AI content. Our suggestion would be not to stress too much about it; as long as you’re obtaining value and are not being misinformed, enjoy the free content, as Google says, regardless of who or what created it! 

Dino Kurbegović is a project coordinator and an investor and technology enthusiast with years of experience in managing complex projects. His journey into content writing began in 2014, covering finance, investing, crypto, technology and complex technical topics.

Subscribe to our newsletter and stay updated !