AI Creative Stylometric Fingerprint Tracker

Domain Presets:

Marker Sensitivity:

Comparative Text Input Samples

Real-Time Local Parsing

● Human Authored Corpus 0 words

● Generative AI Output 0 words

Stylometric Fingerprint Overlay

Radar Analysis

Cadence & Complexity Distribution

Burstiness vs Diversity

Syntactic & Transition Highlights

Transitions N-Gram Repeats

Human Diagnostics

AI Diagnostics

Empirical Metric Evaluation

AI Signature Detected

Stylometric Metric	Human Sample	AI Sample	Delta Difference

Diagnostic Signature Verdict

AI-characteristic structural uniformity detected in clause length variance and transition marker frequency

Text metrics: reproducible counts and detector limits

Your browser does not support video.

Read the explanation

This tracker computes descriptive text statistics. For red red blue blue, four raw words contain two unique cleaned tokens, giving lexical diversity point five. Red blue green yellow has four unique tokens in four words, giving one. Bars share five hundred pixels per ratio unit. Punctuation is stripped from cleaned tokens, while the denominator remains the raw whitespace word count, so unusual tokens can affect interpretation. This type-token ratio depends on sample length and wording. It does not determine whether a person or a language model wrote the text. The human and AI column names describe editor roles, not verified authorship. Sentence lengths are counted after splitting on periods, question marks and exclamation marks. Four words in one sentence give mean length four and standard deviation zero, so burstiness is zero. The same happens for both four-word examples. The source classifies the AI-column text as AI-characteristic when burstiness is below point forty-five or transition density exceeds one. That can flag a perfectly ordinary short sentence written by a person. Both bars are zero because this metric contains no variation for a one-sentence example. No calibrated false-positive rate, provenance evidence or authorship probability is calculated. The threshold merely selects an interface caption. Two equally frequent tokens have Shannon entropy one bit. Four unique tokens occurring once each give two bits. The bars use two hundred forty pixels per bit. Bigram repetition instead compares unique adjacent pairs with the number of pairs; red red blue blue has three distinct pairs, so repetition is zero despite repeated individual words. Transition density searches substring occurrences of a stored marker list rather than a grammatical analysis. The table delta currently reads an undefined property and displays NaN; exports still contain the computed metrics. This explanation preserves that original defect and does not convert these descriptive statistics into a validated AI-text detector.