The words ChatGPT overuses
By Sourabh Singh18 September 20266 min read
Why models reach for these phrases, and why cutting them is the highest-value edit available.
Model output has a recognisable vocabulary. Not because anyone designed one, but because a model picks high-probability continuations, and the highest-probability phrase is, by definition, the one everybody has already written.
The recurring offenders
- Openers: In today's fast-paced world · In the ever-evolving landscape of · It is important to note that
- Transitions: Moreover · Furthermore · Additionally · In conclusion
- Empty verbs: serves as a testament to · plays a crucial role in · underscores the importance of
- Abstractions: navigate the complexities of · a multifaceted approach · leverage a wide range of
- Hedges: it is worth noting · arguably · in many ways
Why they are a double penalty
They are high-probability, which is exactly what low perplexity means, so they push a detection score up. And they are empty, so they cost you nothing to remove. It is rare for one edit to improve both the measurement and the prose, and this is that edit.
What to do instead
Cut most of them outright, a sentence beginning "It is important to note that X" almost always reads better as "X". Replace the abstractions with the specific thing you meant. "Navigate the complexities of hiring" is either "hire fast enough" or "hire without lowering the bar", and which one you meant matters.
The caveat
These phrases are not proof of anything. Human writers use them constantly, particularly in corporate and academic registers where they are practically mandatory. Their presence is a signal about style, not about authorship.
Find every dead phrase in your draft, marked in place.
Scan for clichésFound a paragraph full of them? Rewrite it in your own voice.
Rewrite a paragraphCommon questions
Common ones include openers like "In today's fast-paced world" and "It is important to note that", transitions like Moreover and Furthermore, and empty verbs like "plays a crucial role in". Abstractions such as "navigate the complexities of" and hedges like "it is worth noting" also recur.
A model picks high-probability continuations, and the highest-probability phrase is the one everybody has already written. Nobody designed the vocabulary on purpose.
It can, because these phrases are high-probability, which is what low perplexity means. Cutting them also improves the prose, since they are empty.
No. Human writers use them constantly, particularly in corporate and academic writing. Their presence is a signal about style, not authorship.
