These ChatGPT Words Get On My Nerves.

Minimize using these words, plus how to make your AI-aided content sound less AI.

When ChatGPT was first released, I was super excited like everyone. Being able to receive responses after a query was a game changer for me, who browsed a lot for research or simply out of curiosity.

It was truly a better Google, answering my questions right on the spot instead of giving me a bunch of links to click on.

The language model was initially made available to the public through a free trial. Coupled with its potential for a wide range of applications, it blew up across the internet, reaching 100 million users in just two months.

My classmates started to use it for research. The poetry club in high school I joined for only two weeks used it to generate poems. Even my English teacher approved it for research and brainstorming for all sorts of essays.

Then, there came the problem it had with writing. Students began to copy and paste the whole AI-generated texts into their papers. At most, they would do some tweaks here and there, which were enough to bypass AI detectors.

Of course, such problems are not recent, but my experience with them is, not in school but here across the Internet. As much as I tried to ignore blatantly AI-generated articles on Medium, it’s a pain in my eyes to see how many writers don’t give much effort into making their articles sound less AI.

And such annoyance begins with an introduction like this:

In the fast-paced world of online content creation, efficiency is key. Many aspiring bloggers and content creators find themselves facing the challenge of producing high-quality blog posts within tight time constraints. While writing a compelling and informative blog post may seem like a time-consuming task, it is possible to streamline the process and achieve this goal in just one hour. This essay explores practical tips and strategies to help individuals write a blog post efficiently without compromising on quality. (ChatGPT)

Don’t get me wrong. The introduction is fine in how it is presented. It is well structured; it acknowledges the struggles of producing quality content with tight time constraints. It has a nice professional and formal tone. It gets straight to the point.

What gets on my nerves is this phrase: “In the fast-paced world of online content creation”. Have you seen a similar phrase somewhere before? I’ve seen tons of them:

In an increasingly realistic world…

In the dynamic world of entrepreneurship

In this world for heaven’s sake.

We used to think phrases like these are professional, as they help with framing discussions or introducing topics. I remember the time when students and online writers had to read piles of books, journals, and literature texts for research.

Back in middle school in my country (where technology was limited), I was given sheets of paper to take notes and printed mind maps to organize my ideas. If everything was done as instructed, I would produce an essay so good I could apply to college with it alone.

Now, anything that seems to be a really good piece of writing turned out to be some texts churned out by a bot trained on a massive dataset of repeated texts.

ChatGPT’s most overused words/phrases

Writer Jordan Gibbs on Medium analyzed 1.2 million GPT words to find out the most overused phrases. Out of those words, “can lead to” is the most common GPT three-word phrase. Interestingly, ChatGPT overuses “remember the key” the most, followed by “this could involve” and “here are several”.

You can read more data using this link:

In literature texts, ChatGPT overuses expository phrases like “prepare to embark on” and “Dive into the world of” to introduce plots. Common in the concluding paragraphs are phrases like “as I sit on” and “so let’s”.

In essays, “serves as a reminder” is overused to show lessons learned. Phrases like “it’s important to note that” and “it’s worth mentioning that” are overused to show emphasis. And of course, there’s a myriad of other overused words in typical ChatGPT responses: tapestry, transcend, “foster a sense of”, embrace, etc.

You can see these words for yourself by asking the AI to write five-paragraph essays elaborating on some topic. Things would start to get messy from there.

ChatGPT’s default style: when “average” becomes an annoying shortcoming

Except when you try to steer ChatGPT to a different way of generating text, it has an eerily consistent tone and style to its response from word choice, grammar, vocabulary, and text structure. A person on Linguistics defines this as a result of “the average effect”.

On the bright side, this consistency has given it an edge on tasks like copywriting. But for writing that connects humans, ChatGPT is nowhere near pretending to be a human writer.

A common critique of large language models like this one points to its document retrieval system, which is defined as:

A set of algorithms that find information by matching text records to user queries.

This system prevents true comprehension and critical thinking, as its natural writing style depends on the information sources it is trained on.

GPT-3 is trained on a large collection of text data primarily from the Common Crawl, which includes books, articles, and research papers. Consequently, it naturally leans towards formal writing, maintaining a neutral, somewhat generic tone.

Additionally, this system is prone to statistical bias, where certain words/phrases that appear more frequently in the training data can be overrepresented in the content generated by ChatGPT.

Make Your AI content sound less AI

I believe ChatGPT to be a godsend for many writers, but it doesn’t take away the fact that it is terrible at pretending to be a human writer. Of course, ChatGPT changes regularly, and different models may have different or improved output behaviors.

For me, I have used ChatGPT to brainstorm introductions and conclusions, and it is truly a godsend when it comes to resolving writer’s block. But you can do a whole lot more: generating/organizing ideas, brainstorming, conducting research.

So many writers out there have been using AI to post 1 to 2 articles a day, and I really need them to be careful. If you try to ask it something without being specific, ChatGPT may resort to its internal, neutral tendencies. This is good for idea generation and narrowing down information, but more effort needs to be put in.

Try using notes and mind maps, or any old good way to help you think. For research, I recommend using Google Bard, which in my opinion is more effective in this specific task. But most importantly, read as much as you can.

Ask ChatGPT several questions and read its response to give you options in terms of the direction of your article and its general structure.

Ask more specific questions to layer up that structure. Use the training feature to steer it from its default pattern of discourse. Plan and do multiple drafting sessions, and cut parts that don’t add to your arguments or the overall coherence.

Last but not least, add a human touch to your piece. Tell a story or an anecdote preferably in the introduction. I am an avid Medium reader; when I click your article, I need to know where it is heading.

A story is usually effective because it conveniently establishes the what, who, when, where, why, and how of your piece. Learn more about how to make a story appealing:

“In this era where digital technology” is advancing blazingly quickly, it’s infeasible to tell people to stop using AI. But since the beginning of its creation, AI is destined to be a tool, not a crutch.

For us Medium writers, a post is a story that makes known our position on an issue in a chaotic world, not a recount of information churned out by bots of the latter.

Happy writing!