As AI-generated text proliferates across publishing and media, distinguishing it from human writing has become surprisingly difficult—most people succeed only about 60% of the time. The core problem is that AI systems were trained on human text, while humans now absorb AI language patterns, blurring the line between the two. Commercial detection tools exist but carry serious limitations, particularly for neurodivergent writers or those whose style overlaps with AI output.
Summaries like this, in your inbox every morning.
Sign up free →What happened
As AI-generated text floods publishing, media, and advertising, experts warn that people are remarkably bad at identifying machine writing—a forensic linguist's test shows users succeed only about 60% of the time. Meanwhile, accusations of AI use now routinely bedevil writers (some with justification, others without), and commercial detection tools face significant limitations.
Why it matters
The problem cuts both ways. AI systems train on human writing, while humans are now stylistically influenced by AI, creating what one linguist calls 'a kind of linguistic hall of mirrors.' This makes it nearly impossible to say with certainty whether any individual piece is AI or human-written, unless the author admits it. Detection tools can also produce false results for naturally neurodivergent writers or those who have simply absorbed AI-influenced language patterns.
What to watch
Researchers have identified specific patterns in AI writing—LLMs favor nouns and attributive adjectives, use pronouns less often, and overuse words like 'delve,' 'showcase,' 'boast,' and 'intricate.' However, these same patterns also appear in human writing, including work by Charles Dickens. One detector, Pangram, claims false positive rates of around 1 in 10,000, but independent scrutiny shows its efficacy depends heavily on the register and style choices of the human writer.
No comments yet. Be the first to share your thoughts!
Log in to join the discussion
Get curated AI news from 200+ sources delivered daily to your inbox. Free to use.
Get Started FreeFree · takes 30 seconds · unsubscribe anytime
1 minute a day. The AI essentials.
200+ sources · Email / LINE / Slack
