The Em Dash Didn’t Do Anything Wrong
Somewhere along the way, a punctuation mark older than the typewriter became “proof” of ChatGPT. It’s time to call off the witch hunt.
What we cover
Detectors guess. Humanizers bet. Students pay.
Most AI detectors lean on a simple intuition: machine text is too smooth. Even rhythm, predictable words, sentences that all land at the same length. Humanizers sell the opposite — engineered mess. Paste something in and see the most basic version of that signal for yourself.
How detectors actually workInteractive · Burstiness meter
Sentences—
Avg. words—
Variation—
Reads as—
Section 01
Detectors
How AI detectors work, where they fail, and who trusts them anyway.
See allDetectorsAnalysis
Your Essay Isn’t a Robot. Neither Is the Constitution.
False positives aren’t a rounding error in AI detection. They are the predictable result of how detectors work — and the people they land on are not random.
DetectorsAnalysis
Watermarks Were Supposed to Settle This
Hide a statistical signature in every sentence a model writes, and detection becomes math instead of guesswork. So why hasn’t it ended the argument?
DetectorsHistory
OpenAI Built a Detector, Then Quietly Turned It Off
In 2023, the company behind ChatGPT released its own AI text classifier. Six months later it was gone, citing a “low rate of accuracy.” The lesson still hasn’t sunk in.
Section 02
Humanizers
The tools that promise to make machine text pass as human — and what they really do.
See allHumanizersInvestigation
Inside the Humanizer Economy
Word limits, “undetectable” guarantees and comparison pages written by the vendors themselves. A tour of the business built on beating the detectors.
HumanizersExplainer
Paraphrase, Rinse, Repeat: What a Humanizer Actually Does to Your Sentences
Synonym swaps, sentence surgery, planted typos and invisible characters. A teardown of the techniques humanizers use, and what they cost the writing.
ClassroomAnalysis
The Syllabus Clause Wars
Total bans, open doors and traffic lights: how course AI policies are being written in real time, and why students keep getting caught between them.
LabMethodology
How We Test Detectors and Humanizers
Corpora, thresholds, blind readers and the numbers we refuse to publish without context. The DetectionGuide testing protocol, in full.
Editor’s picks
Start with these