LINGUISTIC FORENSICS · CLIENT-SIDE HEURISTICS
Signal / Noise
Scans your text from the inside out — sentence rhythm, repetition, formality, and AI-typical phrasing. This is a probabilistic estimate, not a guaranteed answer.
Supports English · Hindi · Urdu · Arabic · Bengali · Spanish · French · German · Portuguese — auto-detected.
Input Text
⚠ Too few words — results won’t be reliable. Try at least ~50 words.
Verdict
Human-like
Uncertain
AI-like
Sentence-length rhythm (burstiness) — each bar is the word count of one sentence
—
—
Breakdown — 7 signals behind the score
What this tool does: it measures your text against real language patterns — sentence-length variance (burstiness), phrase-level repetition, AI-typical filler phrases (“in conclusion”, “it’s important to note”, “delve into”), sentence-opener patterns, formality/contraction rate, listing/enumeration patterns, and markdown/listicle structure (headers, bullets, horizontal rules — the typical skeleton of an AI-written SEO article). Everything runs in your browser — no data is sent to any server.
Limitation: this is not a true neural perplexity model (like the ones GPTZero/Turnitin use internally) — it’s a set of statistical/stylometric heuristics drawn from public research. Short text, heavily-edited AI text, or very formal human writing can throw it off. Don’t treat this as sole proof for any high-stakes (academic/legal) decision.
Limitation: this is not a true neural perplexity model (like the ones GPTZero/Turnitin use internally) — it’s a set of statistical/stylometric heuristics drawn from public research. Short text, heavily-edited AI text, or very formal human writing can throw it off. Don’t treat this as sole proof for any high-stakes (academic/legal) decision.