With the rise of chatbots such as ChatGPT, more and more people are warming up to the idea of utilising AI in their everyday lives. From finding new recipes and creating travel itineraries to more complex tasks such as writing code and automating tasks and workflows, ChatGPT has revolutionised how we complete online tasks. 

But what about the more “unethical” tasks, like writing final exams or using AI to pass assessments? Can ChatGPT go as far as helping people graduate from university? Well, yes. But also no. ChatGPT can write well. There’s no denying that. But let’s be honest, most of the time, the writing is bland, unfocused, and ranges from cringe or outright absurd. 

As a writer, I can easily tell when something is written by AI, especially when the user is lazy enough to simply input, “Write me an article about XYZ” or “I need an Instagram caption about XYZ”. However, if you’re someone who doesn’t have a background in writing or editing, you probably ask yourself quite often, did ChatGPT write this? Or, is there a way to detect ChatGPT written content? And the answer is yes, yes, there is. 

In this article, we’ll explain how to detect ChatGPT writing or how to at least make it easier for you to detect.

RELATED: Is ChatGPT Getting Worse Over Time

How to detect ChatGPT writing    

For the sake of this article, I’ve asked ChatGPT-3.5 to write some text. I then break down the text to provide examples of what to look for. 

Example 1 - Introduction paragraph 

The brief

I asked ChatGPT to “write me an introduction paragraph for an article called “How to tell if something was written by ChatGPT”. 

The output

 The verdict

As I said earlier, ChatGPT writes well. However, the above text features overly complex sentence structures such as,

“Understanding the telltale signs that reveal its hand in crafting a composition becomes both an art and a science” 

And an unusually formal tone, 

“We delve deep into the intricacies of language and the subtle nuances that might betray the digital origins of a piece of text”. 

Also, “Join us on this exploration”, you’d think we were taking a trip to Antarctica, not reading about how to detect ChatGPT writing. 

“Fascinating challenge”, I wouldn’t say this topic is fascinating, would you? If what you’re reading features unusual or incorrect wording, that’s also usually a red flag.   

Example 2 - Health article

The brief

I asked ChatGPT to “write me an article about Intermittent Fasting”. 

The output

Firstly, the article it produced was nearly 3,000 words long, which is another telltale sign. AI-generated articles are usually unnecessarily long and wordy. They also don’t always make sense. For example, this is a section from the article:

 
‍The verdict

Benefits and Mechanism? Really, ChatGPT? Unless you’re writing a Thesis, no one talks like this. And to add, if you’re wondering how to detect ChatGPT writing in a Thesis, the lack of credible sources should be a big hint. 

“This eating plan works by prolonging the period when your body has burned through the calories consumed during your last meal and begins burning fat.” Again, this sentence is quite complex and won’t make sense to a general audience. Rule of thumb: if it’s complicated to read and seems to ramble with intricate sentences that don’t really make sense, it’s probably been written by ChatGPT. 

”In summary, intermittent fasting has numerous potential health benefits including longevity promotion, cognitive enhancement, weight management assistance, and disease prevention/reversal”. 

ChatGPT's summary simply repeated the “benefits and mechanisms” listed above without actually summarising the section. So watch out for repetitive words or phrases, as these will likely suggest AI-written content. 

Example 3 - Promotional email 

The brief

I asked ChatGPT to “Write a paragraph for an email promoting a 50% off storewide sale on gluten-free products" 

The output

 
‍The verdict

I think this one’s quite obvious. The text is consistent and fluent but lacks the personal touch of naturally written text. For example, “Dear Health Enthusiast”, I don’t know any company that would address their customers like this. 

”Our gluten-free health products” and “From wholesome gluten-free snacks to nourishing supplements, we've got everything you need to support your healthy lifestyle”. These are vague statements that don’t provide much information. The text reads like a cheesy TV infomercial that’s impersonal and downright cringe. 

”Indulge in guilt-free treats, discover new flavours, and explore a wide array of natural, gluten-free options without breaking the bank. It's our way of making your health journey more affordable and enjoyable. Hurry, stock up on your favourites, and embrace a healthier you with a shopping experience that's light on your wallet but rich in benefits.”  These three sentences are redundant (repeat the same information). This usually happens when ChatGPT isn’t given specific instructions, so it’s using “fillers” to bulk out the text.  

Example 4 - A more in-depth input 

For the first few examples, I kept the brief simple and vague. However, as you may know, ChatGPT’s output is only as good as the input. In other words, give the chatbot more in-depth details, and it should produce better quality content. Let’s give it a go. 

The brief

WeAreDevelopers are Europe’s number 1 developer community, where you can browse developer jobs from companies across Europe. They currently have 2,400+ job opportunities. With a free profile, you can:  

Access career opportunities, tech talks, and events
Get matched with job opportunities on your terms
Gain instant, on-demand access to thousands of hours of tech content
Get exclusive event invites and much more

I need you to write the text for their home page. Approx 500 words. When presenting the information,  use headlines and break text into sections, rather than using a big block of text.

Use a conversational but professional tone. Avoid complexity without talking down to the reader.

The output

The verdict

There are a lot of redundant and complex sentences in the output. Also, ”Dive in now and let your developer dreams take flight.” Perhaps this sentence would make sense if WeAreDevelopers were an aviation school.   

For some reason, nouns like journey, embark, and realm are favoured by AI chatbots. So unless you’re reading about a group of diverse beings uniting to destroy a powerful ring, you should assume the content is AI-written if you see these words. 

Is there a ChatGPT detector?

If you’re still unsure how to detect ChatGPT writing, there are AI detectors on the market which claim to discern whether your text is human or AI-generated - AI detecting AI, oh, the irony. 

However, these AI detectors have been known to provide false positives. For example, one software claimed that The Constitution of the United States was written by AI.  

Most AI-detecting software also only claim to be 98% accurate. If that’s true, that still means one in 50 inputs may still be flagged as being AI-written, even if it’s not. 

So, to further prove the inaccuracy of AI detectors, I tested one by copying the first sentence of this article. The results: (*facepalm)

‍

Key takeaways 

To detect ChatGPT writing, look for the telltale signs such as:

• Unusually formal tone in text that’s meant to be conversational or casual 
• Overly complex sentence structures
• Unusual or incorrect wording
• Text that’s unnecessarily long and wordy 
• Lack of credible sources
• Repetitive words or phrases
• Redundant sentences 
• Lack of personal touch 
• Vague statements that don’t provide much information
• Nouns like journey, embark, and realm

And remember, relying solely on AI detectors could lead to wrongly accusing people, so be cautious and use more than one way to verify before taking any action.