ChatGPT, Claude, Gemini, and the myriad LLM chatbots train on massive datasets. They have access to the sum total of human knowledge, but often get focused on certain words. You may have noticed some of these words yourself, but I'll draw attention to one in particular: delve. Paul Graham, the founder of Y Combinator, drew attention to this with this post.
The use of delve has skyrocketed since the launch of ChatGPT, and while it is a lovely word, it now comes with baggage. If you are pair writing with an AI, you likely have the goal of sounding like a fully human author. If you let ChatGPT choose your vocabulary, you run the risk that it will insert other words like delve. To avoid this, learn to recognize words that these LLMs overuse, and strike them from your writing. This type of process is frequently referred to as "Negative Prompting".
Words
The words you use on a regular basis make up your preferred vocabulary. People get stuck on a word and overuse it, later self correcting to avoid sounding repetitive. ChatGPT unfortunately doesn't adapt this way on its own, so we create word lists to help it along the way. Here is a list of words that you can ask ChatGPT to avoid using.
adventure, alternatively, although, also, architect, arguably, beacon, because, bustling, buckle up, bombastic, captivating, celebrate, comprehensively, conclusion, consequently, crucial, cusp, delve, depicted, demystified, discover, dynamic, elevate, embrace, embark, enable, enrich, ensure, essentially, ever-evolving, explore, excited, fast-changing, folks, fostering, furthermore, game changer, generally, gossamer, groundbreaking, harnessing, headache, highly, hurdles, importantly, increasingly, intricate, integrate, integration, it depends on, labyrinthine, labyrinth, landscape, leverage, linchpin, multifaceted, nestled, namely, navigate, new era, nonetheless, originally, picturesque, pitfalls, poised, pride, profoundly, realm, resonate, robust, revolutionize, seamlessly, seasoned, significantly, similarly, soul, subsequently, supercharge, symphony, tailored, tapestry, testament, thus, top-notch, treasure, treasure box, treasure trove, ultimately, uncharted, undeniably, unexplored, unleash, unprecedented, unravel, unveil, unveiled, vehicle, vibrant, vividly, whispering, whimsical, while, weighing
A recommendation for a ChatGPT prompt for revising content using this word list is below:
The words or phrases listed below are not allowed. If you find any of them in the provided text, please remove and replace them with simpler words that a 2nd grader would use. Ensure the replacements are easy to understand and simple.
Phrases
Single words are a good start to improving your collaborative writing, but it isn't a complete fix. There are also phrases that LLMs love to use, which can also make your writing sound robotic. Here are some phrases and language styles that you can also avoid.
in this article, in this comprehensive guide, here are several, join us as we, embark on a journey, embark on a quest, embark on a fascinating journey, embark on a culinary journey, embark on this gastronomical adventure, this article invites you, this article serves as your, this article acts as your compass, this article will provide you with, explore the art of, master the art of, we invite you to explore, it's essential to, join us in this exploration, within the confines of this article, important to remember, important to consider, worth noting that, stands as a testament to, serve as your roadmap, when it comes to, in the world of, in an era, in a world filled with, in the dynamic landscape of, in the rapidly evolving landscape of, rich tapestry, diverse tapestry of, timeless discipline, the cornerstone upon which, navigating a complex maze, unravelling the mysteries of, captivating exploration, traverse the diverse, unlocking the power of, holds a treasure trove, peel back the layers of this, unfolds like a tapestry of, elegance of artistry, embark on a deep dive, delve into the multifaceted narrative, in conclusion, in summary, to summarize, our journey will provide, whether you’re a seasoned pro, elevate, facets of our lives, navigate the uncharted water, converge to carve a path, embark on this odyssey, when it comes to, a testimony to, cut corners, think outside the box, break the ice, it’s not rocket science, snowball effect, perfect storm, back to the drawing board, raise the bar, get out of hand, go the extra mile, it's important to bear in mind, not only... but also.
If some of these phrases align with how you naturally write, feel free to adapt the list. The reason why this article is here is because everyone is over-conforming to a single style.
Writing Structures
Next, let's delve (I'll bet you noticed that) into some of the writing structures that ChatGPT uses often. One of the most common issue is how it creates titles that use a colon. The title is the hook, the attention getting aspect to your writing. If you immediately signal that you wrote your title with an LLM, you can turn off potential readers. Note that some of the other words and phrases we saw earlier in the article are present in this output below, such as "In an era", "it's essential to", and "consequently".
The structures go deeper than the title, they extend to the entire LLM writing style. They will output a very consistent introduction, body paragraphs, and conclusion. The writing is usually very general in the introduction and conclusion sections, with the quality writing occurring in the body paragraphs. ChatGPT can be militant about following this accurate process of writing. Unfortunately this can cause issues for authors, especially if you chain content together.
One of the best solutions to this issue is to actually treat ChatGPT as an editor. This may be a bit of a letdown, especially for people entranced by the idea of an AI doing all their work. Not all is lost however, there is of course an easy fix. Write a paragraph yourself, then ask ChatGPT to improve it or take it to the next level. When you've finished a rough body of an article through this process, you can then ask ChatGPT to write an outstanding introduction or conclusion.
Also take into account that ChatGPT truly loves lists and will find a way to work them into practically any content. Here is an example from a prompt about sandwiches.
Whether this is pop journalism's fault for recognizing and abusing the general public's weakness for lists of "5 Things You Didn't Know About Kim Kardashian" (I'll have you know that I wasn't aware that was a real article when I wrote this), or simply a good way to visually break up blocks of text, the fact is ChatGPT powered lists are here to stay.
However, I strongly encourage you to not simply accept the first list thrown at you by an LLM, because they frequently can do better if compelled to. Don't feel the need to include a full list of five bullets if two or three options will do. I'll conclude this with a famous story about the sculptor Michelangelo.
Michelangelo Quote:
This famous quote about the creation of the Statue of David is often attributed to the artist.
“The sculpture is already complete within the marble block, before I start my work. It is already there, I just have to chisel away the superfluous material.”
ChatGPT and the other LLMs have given every person on earth access to incredible generative abilities to build "AI Marble" for essentially no cost. These AI chatbots are capable of writing more text in a single day than the sum total of human writing to this moment. Understand that when you use them to create work, you must also be conscious of what not to write. The purpose of writing is to be read by someone, and if you alienate your audience by making them wonder about the authenticity of your work, nobody involved will be pleased.
