Paste any text. Highlights the 147 phrases that flag content as AI-written. Based on the Kompozy AI Tells Index.
Paste any AI-generated text. We'll flag the 50+ AI-tell words and phrases that scream "ChatGPT wrote this." Higher human score = sounds more like you.
A word or phrase that LLMs over-use vs human writers — "delve", "leverage", "in today's fast-paced world", "navigate the complexities", "tapestry". Readers spot 3+ tells per paragraph and start skimming or close the tab. The list is built from analysis of 50k+ ChatGPT and Claude outputs vs human-written content in the same niches.
Tells per word, weighted. Score = 100 - (tells/words × 600), floored at 0. A 200-word paragraph with 3 tells lands ~91. Same paragraph with 7 tells lands ~79 (editing required). 12+ tells lands under 65 (reads as AI). Custom banned words count toward the same score.
Not directly — it detects the surface patterns LLMs lean on. Human writers occasionally use these words too. The score is a writing-quality signal, not a detector. For actual AI detection, use Originality.ai or GPTZero (less reliable than they claim).
Replace abstract verbs ("leverage" → "use"), kill filler transitions ("furthermore", "moreover"), drop hedging ("it's important to note"), prefer concrete nouns over metaphor ("tapestry of strategies" → "the 4 plays we ran"), and read the draft aloud — anything that sounds like a LinkedIn keynote needs to be rewritten.
Kompozy automates everything this tool estimates — multi-format generation, brand voice, scheduling across 9 platforms. Free 500-credit trial, no card required.