Try free online ChatGPT detector
Try free online Keyword Extractor and Text Analyzer
Since 2023, ChatGPT and other AI tools capable of generating text have made an enormous impact on how content is created. As these tools are reshaping the web, we are witnessing how the development in AI contributes to an ever increasing mass of junk posts not only across the web, but also in schools, academia and news agencies, not to mention digital marketing and SEO.
AI-written texts on spam websites that are professionally optimized for search engine indexing are able to receive better search positions when compared to authentic posts that were originally written by humans. This seriously undermines search ranking of sites with legitimate content. The vulnerability comes from LLMs’ ability to instantly rewrite authentic texts in a way that is indistinguishable by search engines. As as a result, the web is flooded with posts that are slightly different versions of same text, but do not literally reproduced the original.
Certain types AI-generated content is useless, uninformative, misleading; most all of it lacks authenticity. This is why there should be a way for anyone to quickly discern if dealing with posts written by humans or a LLM. To address the necessity, we have created and implemented a vocabulary-based ChatGPT detector.
You can copy and paste the text that you want to test and instantly get the probability if it was written by ChatGPT.
You can try the ChatGPT Detector here. The client-side tester will scan the text and give a quick cheat sheet reference of words and phrases that resemble the language of an AI-generated post. It works with English and takes less then a second to process any input. We are constantly updating the vocabulary test to keep it up-to-date with the progress of ChatGPT. Unlike other similar tools, our detector is very easy to implement as it transparently uses the method of bigrams, otherwise tedious to apply with a simple «ChatGPT cheat sheet».
Our detector relies on a list of ngrams that were extracted from an English corpus of ChatGPT replies. We found that the bulk of posts that were composed by ChatGPT uses predictive words and phrases. They are easily captured by human eye if you are used to see replies to prompts and can be statistically measured. If the number of such vocabulary markers exceeds a certain threshold, that would suggest that the text is likely composed by a LLM. You can read our preliminary research on this topic in the article, «The Intricate Tapestry of ChatGPT Texts: Why LLM overuses some words at the expense of others?». The phenomenon of vocabulary overuse (and underuse) by LLMs relates to the idea that during the training of language models like GPT, the model is optimized to minimize a loss function which measures the difference between the model’s predictions and the ground truth during supervised learning.
Go to ChatGPT detector.