AI Text Stylometry Detector

An analysis tool that scores, on a per-sample and heuristic basis, how likely a text corpus (tweet timeline, LinkedIn posts, blog articles, email archive) may have been AI-generated. It does not use deep ML models (DetectGPT, GPTZero); instead it works across 6 classic stylometry axes: sentence-length variance (burstiness), lexical diversity (TTR), generic/filler phrase density, sentence-opener repetition, em-dash and punctuation overuse, and posting-rhythm anomaly.

When to use: When you want to audit whether an account's or author's content is organic or sourced from an AI-assistant/automation; for competitor content audits, bot/pipeline detection, or the "did GPT write this text" question.

Output: For each sample, an AI-probability score from 0–1, a 6-axis rubric, a classification (likely-human / mixed / possibly AI-assisted / likely AI-generated), evidence strength (low/medium/high/conclusive), and a list of supporting axes. For the corpus overall it produces an aggregate score + axis summary. With an optional baseline corpus, it compares against the author's past "human" style.

Important: The score is a probability, not a certainty. No "AI-generated" claim is made without ≥2 independent axes + contextual evidence; it is calibrated for both English and Turkish.