Paste plain text or LaTeX and get a calibrated AI-likelihood score with signal breakdown and sentence heatmap. Runs entirely in your browser — nothing is uploaded.
Stock phrasing + low burstiness pulled the score up. Confidence 71%.
Furthermore, a robust framework for data collection is essential. Moreover, researchers must delve into the multifaceted nature…
Built for researchers who want a fast originality / AI-plagiarism check before submitting — without creating an account or uploading a manuscript.
Plain prose or LaTeX. Math, commands, and verbatim blocks are stripped so only the writing is scored.
Surprisal, burstiness, lexical diversity, n-grams, function words, openers, punctuation, stock phrases, and contractions.
Gauge + per-signal bars + sentence heatmap. Hover any sentence for its local AI-like score.
Low, flat surprisal across tokens often reads as AI. We use a bundled Zipf frequency model — a proxy, not a full language model.
Humans vary sentence length; many models don’t. Uniform commas and sparse questions or dashes also nudge the score.
“It is important to note,” “delve into,” “multifaceted,” repeated transition openers — classic tells we weight explicitly.
Grammar-word share, moving-average type-token ratio, and contraction rate help catch formal, over-smoothed prose.
Formal human academic writing can score high. Lightly edited AI can score low. No detector, including paid ones, can prove authorship. Use this to revise your own writing — never to accuse someone.
Free, no signup, nothing uploaded. Paste ~60+ words for a reliable estimate — then rewrite the flagged sentences before you submit.