Trust & truth · the prose tells
Spotting AI-written text
30-second gist~30s read
You can sometimes spot AI-written text from its tone — too polished, oddly bland, weirdly even. You can almost never prove it. AI-detectors miss real cases, accuse innocent humans, and fail entire languages. Use signals, not certainty.
If you want more
Three softer signals to notice
- Even paragraphs. Each paragraph the same length. Each sentence a similar shape. Real writing varies a lot more.
- "Three things to consider…" AI loves a structured list, even where one isn't needed. Bullets where prose would do.
- Frictionless transitions. "However…", "Moreover…", "It is important to note that…" — generic connectors used more than a real writer would.
Why AI-detectors are unreliable
Detectors look for statistical patterns ("perplexity") that AI text tends to have. But AI keeps changing, and humans who write neatly are flagged false-positive constantly — particularly people writing in their second language. In 2023 a Stanford study found a leading detector flagged real essays by non-native English students as AI-written 61% of the time. Most universities have quietly stopped relying on detectors.