Artificial intelligence
🔍 AI Detection: Spotting Machine-Written Words and Text
How AI text detectors work, from word-choice tells and perplexity scores to trained classifiers, watermarks and content credentials, how accurate they really are, who gets wrongly flagged, what the law says, and how to judge a claim that something was written by a machine.
- 1What AI Detection Is and Why People CareWhat it means to detect machine-written text, who depends on getting it right, and the documented cases where AI text slipped through and where people were wrongly accused.20 min
- 2How Machines Write, and Why It Leaves FingerprintsHow a language model turns probabilities into sentences, why that process leaves statistical traces, and what perplexity and burstiness measure, with a worked calculation.18 min
- 3How AI Text Detectors WorkThe five families of AI text detectors, what each one actually measures, how the major commercial tools describe themselves, what independent tests found, and how to read the score a detector gives you.20 min
- 4Watermarks and Provenance: Marking Machine Text at the SourceHow AI companies hide statistical watermarks in generated text, why editing, translation and short passages weaken them, how content credentials and platform labels work for images, audio and video, and what changed when the EU marking rules took effect in August 2026.18 min
- 5The Tells: Word Choice, Habits and Invented FactsWhat the research actually shows about AI vocabulary and style tells, why a tell that is real across millions of documents is weak evidence about one document, and why invented citations and quotes are the strongest tell because anyone can check them.18 min
- 6Accuracy and False Positives: Who Gets Flagged and WhyHow detector accuracy is measured, why a 1 percent false positive rate still flags hundreds of innocent people at scale, what independent tests found, which writers get wrongly flagged most, how paraphrasing and humanizer tools defeat detectors, and which universities turned their detectors off.19 min
- 7Detection in Practice: How Institutions Handle Suspected AI TextWhat schools, universities, newsrooms, journals, search engines, marketplaces, platforms and employers actually do when they suspect machine-written text, and why the careful ones rely on process and records more than on detector scores.17 min
- 8Law and Policy: Rules on AI-Generated Text and Its LabelsWhat the EU AI Act's Article 50, the FTC's fake review rule and enforcement actions, the U.S. Copyright Office, Thaler v. Perlmutter, Idaho's election and chatbot laws and California's AI Transparency Act actually require about machine-written text, who each one covers and when it applies.18 min
- 9Judging a Claim That Something Was Written by AIA step-by-step method for weighing an accusation that a text was machine-written, the kinds of evidence that settle authorship and the kinds that do not, how officials can handle a flood of public comments, and how writers can keep proof of their own work.17 min