🔬
How it works
Analyses perplexity (how surprising word choices are) and burstiness (variation in sentence length) — the two key patterns that differ between AI and human writing.
🎯
Sentence highlighting
Each sentence is individually scored. Pink highlights indicate AI-like patterns, green indicates more natural human writing variation.
⚡
Instant & private
Analysis runs entirely in your browser. Your text is never sent to any server or stored anywhere.
📚
Who uses this
Teachers checking student work, editors reviewing submissions, HR screening applications, and content managers verifying original writing.
Frequently asked questions
How accurate is this AI checker?
This tool uses statistical analysis of perplexity and burstiness — the same core signals used by professional AI detectors. Results are most accurate on longer texts (200+ words). Short texts under 50 words may give unreliable results. No AI detector is 100% accurate — always use results as one signal, not a final verdict.
Is my text stored or logged?
No. All analysis happens in your browser using JavaScript. Your text is never sent to our servers, never stored, and never logged. It disappears when you close the tab.
What is perplexity in AI detection?
Perplexity measures how predictable the word choices are. AI models tend to choose the most statistically likely next word, resulting in low perplexity (very predictable text). Humans make more surprising, varied word choices, resulting in higher perplexity. A consistently low perplexity score is a strong signal of AI-generated text.
What is burstiness?
Burstiness measures variation in sentence length. Human writers naturally vary their sentences — some very short, some very long. AI tends to produce sentences of similar, moderate length. Low burstiness (uniform sentence lengths) is a strong indicator of AI writing.
Can it detect text that was edited after AI generation?
Partially. If someone makes minor edits to AI text, the core statistical patterns remain. Heavy editing or running text through a paraphrasing tool can reduce the AI signal. Scores above 60% are still worth investigating even on edited text.
Which AI models can it detect?
The checker works against patterns common to all large language models — ChatGPT (GPT-4, GPT-3.5), Claude, Gemini, Llama, Mistral, and others. It detects the statistical fingerprint of AI writing rather than signatures specific to one model.