AI text detector
Estimate how likely a text is to be AI-written, from local writing-style heuristics.
AI text detector
Paste any passage and this tool estimates how likely it is to be AI-written, using a handful of local writing-style heuristics — no text is uploaded anywhere, and no external AI-detection API is called. It looks at sentence-length variation (burstiness), how often the text leans on phrases AI chat models overuse ("delve into", "it's important to note that", "a testament to"…), how repetitive its vocabulary is, how uniform its paragraph lengths are, and how much it relies on em dashes and three-item lists. Each signal is reported separately with its own score and a plain-language explanation, so you can see exactly what drove the overall number rather than trusting a single opaque percentage.
The Sensitivity option controls how aggressively the overall score reacts to the signals it finds — Low is conservative and only flags strong, unambiguous patterns, while High reacts more readily to the same signals. Turn on "List flagged phrases" to see every matched phrase from the built-in list, with its line number, printed under the score breakdown; turn it off for a shorter report with just the numbers.
Every signal degrades gracefully on short input: a single sentence has nothing to say about sentence-length variation, so that signal reports "not enough data" rather than guessing from one data point. This matters because none of these signals, alone or combined, is proof — human writers can write in a uniform, phrase-heavy style, and AI output keeps changing to sound more like careful human prose. Treat the score as a prompt to look closer, never as a verdict, and never as the sole basis for accusing someone of using AI.
Everything runs locally in your browser using plain JavaScript — your text is never uploaded or sent to any server. Copy the report, download it as a .txt file, or send the text on to another tool like the AI phrase remover to clean up what this tool flagged.