Estimate whether English text was written by an AI using a combination of three classic heuristics: sentence burstiness, AI-phrase density, and vocabulary uniformity. Returns a 0–100 score with verdict.
AI-generated text has detectable patterns: sentences are uniformly long (low burstiness), it overuses certain phrases ('it's important to note', 'navigating the complexities of'), and its vocabulary spread is more even than typical human writing. Combining these signals gives a probabilistic detection score.
Paste at least 3 sentences. The tool analyses sentence-length variation (humans vary widely; AI is uniform), counts AI-style phrases from a curated list, and measures vocabulary diversity. The three signals are weighted into a 0–100 score with a human-readable verdict.
Use it when reviewing freelance writer submissions, when checking student essays, when verifying that LinkedIn or blog content matches the author's usual voice, and when fact-checking suspicious 'expert' content.
This is heuristic-based — for high-stakes detection use GPTZero, Turnitin AI, or similar ML-based services. The tool gives directionally correct results but isn't perfect. Combine with manual reading for confidence.
Roughly 70–80% accurate on typical AI vs human writing. Skilled human writers can score 'AI'; well-edited AI can score 'human'.
The same tool — ChatGPT writing trips the same heuristics. There's a separate ChatGPT Detector that uses identical engine, branded for that use case.
Modern LLMs can match human writing very well; reliable detection requires the LLM provider's cooperation (e.g., watermarking).
Explore more plagiarism & ai detection on the tool hub — or jump straight to the Plagiarism Checker, ChatGPT Detector.