Stylometric fingerprints

The habits that make machine-written prose recognisable without any hidden character at all: uniform paragraph rhythm, three-part lists, a fondness for particular constructions, and predictability that a statistical model can measure.

How it works

Stylometry measures features of writing that authors do not consciously control: sentence length distribution, function word frequencies, punctuation habits, and the predictability of each word given the ones before it. A detector typically scores the last of these directly, computing how surprised a language model is by the text, since machine-generated prose sits closer to the model's own expectations than human writing does. It then combines that with simpler features, of which punctuation is the easiest, and produces a probability. There is no code point to remove because the signal is in the choices, not in the characters. This also explains the failure mode that matters ethically: a careful writer with consistent habits, or someone writing in a second language with a limited construction repertoire, produces text that is genuinely more predictable, and the detector cannot tell that apart from generation. Cleaning invisible characters removes one category of evidence, the objective one you never chose. It does not touch this, and no honest tool claims otherwise.

What it is legitimately for

Stylometry is a real discipline, used in authorship attribution and forensic linguistics long before language models existed.

What it breaks

• Nothing technical. This is the part that gets writing flagged after every invisible character has been removed.
• It also flags people. A careful writer with consistent habits looks statistically similar to a machine, which is why detector false positives fall hardest on non-native speakers and on anyone with a tidy style.

How to find it yourself

Read it. If every paragraph is the same length and every list has three items, that is the fingerprint.

What Clipboard Sanitizer does

Nothing. There is no code point to remove. We mention it because a page that promised clean text without saying this would be lying by omission: cleaning characters does not make writing look human, and it was never supposed to.