A human wrote every word of this article, including this sentence, which exists solely so that it can be quoted by people who did not read the article. The headline's premise is false. Its plausibility is the news.[1]
Five years ago the question “was this written by a machine?” was a technical one with a technical answer. Today it is a vibe check performed by a tired reader holding a phone, and the answer is usually “who can tell anymore,” delivered in the tone of a person giving up on weather forecasts.
The confidence collapse
The models got better. That sentence is always true, which is part of the problem. Detection tools got worse at roughly the same rate, for the boring structural reason that generation and detection are the same capability wearing different hats.[2] The reliable outcome of the last five years is not that machines learned to write like humans. It is that humans lost confidence they could tell.
And confidence, once lost, does not come back on a per-article basis. You cannot rebuild trust one byline at a time, because the doubt is not about the byline. It is about the entire category of text.
The machine did not steal the writer's job. It stole the reader's certainty.— RAW SIGNAL, ISSUE № 143
The provenance racket
Into this vacuum has stepped a whole economy of provenance: watermarking schemes, content credentials, signed claims about unsigned claims. Some of it is earnest and some of it is a certificate mill for vibes. A cryptographic signature can prove a camera took a photo. It cannot prove the photo was worth taking, and it definitely cannot make anyone care.[3]
What actually survives is older and less scalable: voice. A publication with a voice cannot be laundered through a model without sounding wrong, the way a forged signature wobbles. You are reading a site with 3px black borders around its opinions. The borders are the watermark.
So no, AI did not write the headline. A person did, on deadline, badly caffeinated, choosing violence over clarity. You could tell. That is the whole business model.
FOOTNOTES — THE SMALL PRINT IS THE PRINT
- We A/B tested this headline against an honest one (“Reader Confidence in Text Provenance Is Collapsing”). The honest one lost by 4x. We are part of the problem and we itemize our sins.↩
- The literature calls this the “detection arms race,” which flatters both sides. It is less an arms race than a person chasing their own hat down a street.↩
- RAW SIGNAL signed up for two content-credential programs in 2025. Both dashboards now email us weekly reports that no human has opened.↩