ChatGPT, the artificial intellect we’ve all come to rely on for everything from homework help to brainstorming has a vocabulary that is both vast and finely tuned, thanks to a meticulous process that blends linguistic analysis with cutting-edge technology. However ChatGPT appears to repeat some terms more often than others which are used to train AI Detectors to recognize ai generated text.

ChatGPT Vocabulary

ChatGPT’s vocabulary is a curated list, initially built from the 245 different UTF-8 symbols, ensuring it can express itself across languages and symbols with ease. This basic set includes everything from the letters of the English alphabet to Chinese characters and even special symbols like “©”. However, it’s the expansion of this vocabulary that truly showcases the model’s linguistic prowess​​.

To enrich its lexicon, OpenAI analyzed millions of webpages, extracting the most frequently occurring character groups. This data, likely derived from a massive trove of content linked from Reddit posts, forms the bedrock of ChatGPT’s understanding of language. By continuously combining common character pairs into new tokens, the model’s vocabulary grew to encompass 100,256 unique tokens, including terms, parts of words, and even popular combinations in programming languages​​.

So, how what to do to make ChatGPT generate text with a less repetitive vocabulary that doesn’t sound AI generated text and protect yourself against AI Detection solutions?

ChatGPT most commonly used words

Interestingly, the process also unveiled phrases that are particularly favored by ChatGPT, distinguishing it from human writing. For instance, it has a penchant for phrases emphasizing importance (“it is important to…”), expressing uncertainty (“can vary depending on”), and presenting multiple options (“its also important to…”). This characteristic style is a direct reflection of the data and human preferences it was trained on​​.

Which Words Does ChatGPT Use the Most?

This insight into ChatGPT’s common list of words and phrases isn’t just an academic exercise. It sheds light on how the model might be “thinking” and the linguistic patterns it prefers, offering a fascinating glimpse into the world of AI communication.

AI Detectors and Humanizers

Interestingly, software houses have started to build a category of tools designed to detect AI-written text by identifying the top repetitive AI generated words to label human-created content and demote AI generate content from search engine negatively affecting your WebSite SEO.

A secondary category of tools, named Humanizers, the do exactly the opposite: the take a text in input, they automatically remove the most commonly ChatGPT used words from it and ask back an LLM to paraphrase it to reduce the probability the text would be detected by an AI Detector tool.

In alternative to using an AI generated text Humaniser, in order to remove these words manually you can also ask ChatGPT to use synonyms in the prompt or simply get a old good thesaurus (a book of words and their synonyms) to find the right replacement to the words I recommended you to avoid.

So, the next time you interact with ChatGPT, or a re reading a web page and you may be wondering if it has been generated by a ChatGPT Based CopyWriting tool, you’re not just engaging with a black box of algorithms; you’re conversing with a system that has a unique linguistic identity shaped by its training. Whether you’re asking for writing advice, coding help, or just curious questions, remember that you’re navigating a carefully constructed linguistic maze, one token at a time entirely constructed on probability that predetermine how likely each word is likely to appear and be repeated again and again and again.