ESL-safe mode active. Common-vocabulary penalty disabled, phrase weighting halved, "Likely AI" threshold raised from 75% → 85%. Designed to reduce false positives on writing by non-native English speakers.
Analyzing your text...
Running statistical signal analysis + LLM signature matching against ~140 patterns.
0%
AI Probability
—
—
Human Words
—
AI-pattern Words
—
Sentences
—
Characters
Most Likely Source
Top match:——
Attribution is based on phrase patterns observed disproportionately in each LLM's output. Heavily edited or paraphrased AI text may not match any signature cleanly.
Sentence-level Analysis
Likely humanPossibly AIStrong AI signal
Detection Signals
AI Phrases /1k words
—
Burstiness Score (AI%)
—
Sentence Diversity (AI%)
—
Punctuation Variety (AI%)
—
Common-word Reliance
—
Paragraph Uniformity
—
AI Crutch Phrases Detected
Detected as AI? Rewrite this text to read as human-written — meaning preserved, AI tells removed.
How accurate is this? No AI detector is 100% reliable. This tool combines statistical signals (burstiness, lexical patterns, structural uniformity, model-specific phrase tells) and attributes likely source by matching against curated phrase libraries for ChatGPT, Claude, Gemini, and generic LLM patterns. Use the result as a hint, not a verdict — heavily edited AI text, short samples, or modern LLM output that's evolved past these signatures may be misclassified. If you're a non-native English speaker, enable ESL-safe mode above to reduce false positives.
