Multi‑model resistant · Anti‑bias · Circular AI/Human proportion · No fake percentages
Upload PDF, DOCX, or TXT (max 5MB)
No file selectedOur engine evaluates 11 independent linguistic dimensions (burstiness, perplexity, transition density, lexical diversity, repetition fingerprint, emotional variability, personal perspective markers, logical optimization, stopword ratio, sentence start variety, and hesitation patterns). Each dimension is scored and weighted – but no single feature triggers an AI accusation. The final proportion (AI / Human) is visualized in a circular donut chart.
Why no simple percentage? We provide an estimated proportion based on the strength of human vs. AI indicators, but if the overall confidence is below 60%, the result is “Uncertain / Inconclusive”. This avoids false accusations against professional or non‑native writers.
Robustness: The detector is calibrated against modern LLMs (GPT‑4, Gemini, Claude 3, Llama 3) and resists paraphrasing, humanized outputs, and mixed human‑AI collaboration.