---
title: "Telling AI-written text from human writing, the measurable differences"
language: en
topics: ["AI", "SEO", "art", "creativity", "data", "people", "writing"]
full_article_url: https://willem.com/en/2024-11-11_write-your-own-words/
---
# Telling AI-written text from human writing, the measurable differences

*Errors and vocabulary give the origin away.*

> Look at errors and vocabulary. In research I cited in 2024, 78% of human-written essays contained at least one mistake against 13% of AI essays, and ChatGPT used 35% less unique vocabulary than human writers.

Researchers from Preply analysed 12,000 student essays to find measurable differences between human and computer-generated writing. Human texts were far more likely to contain mistakes: 78% had at least one error, against only 13% of the AI essays. The most common punctuation errors in human texts were missing commas, hyphen errors and missing spaces. ChatGPT also used 35% less unique vocabulary, 7,308 unique words overall against 11,248 in the human-written texts, and it favoured complex, specialised terms like cultural or economic where humans pick common words like one or people. My wider point stands next to the numbers: if everyone generates content with the same tools, everything starts to look the same, which is why I argue for writing your own words.