ChatGPT is changing how we write – why that can be a problem

Everyone probably knows this: using ChatGPT to write an email that has been putting off for days. The AI helps with formulations and you don’t have to worry about it yourself. But have you ever asked yourself to what extent the large language models (LLM = Large Language Models) have their own writing style? With this, Ritesh Chugh has chosen The Conversation explained.

Artificial intelligence has a penchant for phrases

Have you ever noticed that certain phrases appear over and over again in social media posts, news articles, and even academic texts? For example, phrases like “delve into” (= go into depth) and “navigate the landscape” (= navigate through the landscape, but also in a figurative sense: find your way around a subject area)?

Chugh, who researches information and communication technologies at an Australian university, explains this trend with the increasing use of generative tools such as ChatGPT. The use is understandable: scientists can express themselves in a technically correct manner. Language AI and its formulation aids can be useful so that people who are unfamiliar with the subject do not stumble too much over the wording in scientific papers.

A little digression: The Voice AI learns from human data (texts, conversations, etc.) that is made available to the software. Artificial intelligence learns idioms or phrases that we often use. In addition, she makes decisions based on learned experiences as to which constellations of words and formulations are better, for example when writing an essay or in an email.

“Since the release of ChatGPT, the use of words such as ‘deepen,’ ‘present,’ ‘underline,’ ‘crucial,’ ‘range,’ and ‘meticulous’ in academic texts has increased dramatically,” writes Chugh. So AI tends to use common words in addition to phrases more frequently.

Why ChatGPT’s phrases are a problem – and what you can do

ChatGPT and similar AI models use common terms and popular word combinations as formulation aids, which in itself can be a good thing. But if all the texts are suddenly written in a similar style, “it takes on a robotic undertone,” says Chugh.

This also applies to the style of the language AI: If the text help is too flowery or particularly elaborate, ideas can sometimes not be conveyed as effectively as with clear and straightforward language. One study even came to the conclusion that simple and precise language can not only help with understanding, but also make the author appear more intelligent.

Overuse of certain words can make text boring instead of enhancing it. According to Chugh, there is currently no accurate list of which words ChatGPT prefers to use. In order to avoid ChatGPT and its alternatives from overshooting the mark (to stay with the phrases), the scientist lists the following possible solutions:

1. When using AI tools, pay attention to word repetitions and replace them. Simpler alternatives can often help here: instead of “deepen” you could use “explore”, for example.
2. Give the AI ​​written instructions to use clear and simple language.
3. ChatGPT is not a miracle cure. Edit your text afterwards and change paragraphs that seem strange to you. You can find synonyms using tools like thesaurus.
4. The AI ​​settings are often customizable. With Microsoft Copilot or ChatGPT you can often focus on clarity and simple expressions. Sometimes you can also create lists of words that the AI ​​should no longer use.