Project History
Timeline of detection algorithm development, feature additions, and stylometric research publications.
May 22, 2026
- — Launch of AI Detector. Editors, teachers, and students can now freely analyze texts for traces of machine generation with even higher accuracy.
May 13, 2026
- — Traffic surpassed 2500+ checks per day. Most of it, it turns out, comes from academia: teachers and students are massively checking essays, research papers, and theses.
- — Added answers to frequently asked questions — for example, why pasting text is safer than uploading a DOCX file: that way hidden characters left by AI tools do not end up in the check.
May 8, 2026
- — Traffic surpassed 500+ checks per day. The stats show a massive influx of users checking essays and research papers.
- — Benefit for students: Rolled out a core micro-update. We slightly "loosened the screws" for dry academic styles (the Depersonalization pattern) so the detector is less likely to confuse a tired student writing strictly to academic standards with ChatGPT. Breathe out, fewer false positives. Servers are under load but holding up perfectly.
May 5, 2026
- — Ran our own old articles and press releases from 2022 through the detector. The algorithm mercilessly flagged them as 80% AI. We cried a little, felt ashamed, but admitted it — we really used to write like robots. Also tweaked the logic for detecting "corporate bureaucratese" along the way.
- — Rewrote the names of the signals in the methodology so they are easier to understand.
April 12, 2026
- — Launch of the international version of the service (Orhuman.com) with stylometric analysis adapted for the English language.
March 16, 2026
- — Published research on algorithmic blind spots. Added detailed analysis of formal genres: encyclopedic articles and legal documents.
- — Implemented a visual indicator of the degree of algorithmic text automation in the detector interface.
March 13, 2026
- — Transitioned to the 'Authorship Axis' system. Instead of searching for binary probability, the algorithm now classifies text into 4 categories: authorial, editorial, generated, and hybrid.
- — Integrated a search fact-checking module to verify the authenticity of editorial materials and official documents.
- — Changed display logic: the system now shows detected machine generation patterns even for texts identified as formal documents to ensure analysis transparency.
March 12, 2026
- — The pattern database was expanded with the Falsification section. The algorithm learned to detect deliberate attempts to imitate human style (artificial typos, excessive slang, pseudo-doubts).
- — Published a document regarding the limits of capabilities and technological constraints of the detector.
March 11, 2026
- — Added estimated time of arrival (ETA) calculation during deep text analysis.
- — Updated the methodology section: added visual comparisons of machine and human approaches to text formation.
March 10, 2026
- — Introduced paragraph-by-paragraph analysis: the detector now visually highlights generated and human-written fragments within a single material.
- — Added a feature for local storage of check history. Data is saved exclusively in the user's browser (LocalStorage).
- — Expanded the detector interface: added functions for quick clipboard pasting and copying results.
March 09, 2026
- — Public launch of the project.
- — Published the complete open methodology of stylometric analysis, including a matrix of linguistic and structural markers.