From the Conch team
Why ChatGPT uses the same 200 words to write everything
Language models are probability machines. When generating the next word, they always pick the safest, most expected choice — which is usually one of the same recurring words: "significant", "important", "demonstrate", "challenge", "provide". The result is prose that is grammatically flawless but rhythmically dead. Shannon entropy, the measure this tool uses, was originally developed to measure information in telecommunications signals. Applied to writing, it captures how "surprising" your word choices are. A low entropy score means your writing is highly predictable — the same words keep appearing. Human writers are naturally more erratic, jumping between concrete and abstract, technical and colloquial, precise and poetic. This tool finds where you are stuck in vocabulary loops and shows you which specific words to swap.
Conch Tip: The 'Top Repeated Words' list is your shortcut. Find the top 3 non-stopwords and replace 50% of their occurrences with specific synonyms. That single change usually adds 8–12 entropy points.