InkTrace

Extracts article text, estimates its AI writing characteristics with TypeSafe Jev, and flags how uncertain the result is.

InkTrace checks whether a web article reads as AI-generated. It extracts the article text, asks TypeSafe Jev to analyze its writing, and shows the answer as a model estimate together with how uncertain it is. How it works • Manual detection: on any article, click the InkTrace icon in the toolbar, then click “Detect”. • Auto-detection: on sites you read often, click “Auto-detect on this site” once and allow it in Chrome's prompt. Articles on that site are then checked when you open them. You can pause or revoke any site in the settings. What you see • A verdict: Likely human, Likely AI, Mixed, or Uncertain, with a risk level. • The AI likelihood (a model estimate) and the model's confidence, shown separately. • The writing type, the text length, and how much of the text was analyzed. • When the evidence is weak, the model is unsure, or its estimates disagree, InkTrace says “Uncertain” instead of guessing. Errors are shown as errors, never as a verdict. Requirements • A TypeSafe Jev API key (create one at console.typesafe.ai/keys). Jev bills by input tokens, so the cost of a check depends on the article length. • Chinese and English articles. Text must be at least 400 Chinese characters or 200 English words. Privacy • Only the article title, language and text are sent, and only to TypeSafe Jev (api.typesafe.ai). Never URLs, cookies, account data or browsing history. • Article text is kept in memory only while a check runs. The cache stores analysis results, not text. • No analytics, no telemetry, no other network requests. • Website access is requested per site, only when you turn on auto-detection for it. Limitations AI text detection is uncertain. A result is a model estimate, not proof of who wrote an article or how. Detection is less accurate for Chinese than for English, and AI text that a person has heavily edited may show as Mixed or Uncertain.