AI Text Detector

Five categories: (1) AI vocabulary — over 60 words and phrases that appear disproportionately in LLM output such as "delve", "tapestry", "foster", "underscore", "paradigm". (2) Sentence length variance — AI tends to produce sentences of similar length; humans naturally mix very short and very long sentences. (3) Transition phrases — "it is worth noting", "in conclusion", "having said that" and similar constructions. (4) Style patterns — em dash overuse, "not only…but" constructions, excessive superlatives. (5) Passive voice ratio above natural human rates.

Yes. Lightly editing AI text — replacing flagged words, varying sentence lengths, removing transition phrases — can lower the score significantly. Conversely, a human who naturally uses formal or academic language may score higher than expected. This is an inherent limitation of pattern-based detection. The tool is useful as a quick first check, not as definitive proof.

The vocabulary and transition phrase lists are English-only. Sentence variance and passive voice detection work on any language but are less calibrated. For non-English text, treat the score with extra caution.

No. All analysis runs entirely in your browser with JavaScript. Nothing leaves your device. This makes it safe to analyze confidential documents, unpublished work or sensitive content.

Scores above 60 suggest likely AI origin. Scores above 75 are a strong signal. Scores between 40 and 60 indicate mixed signals — the text may be AI-assisted or lightly edited. Below 40 is consistent with human writing. Short texts (under 100 words) produce less reliable results because there is less signal to detect.

Academic, legal and formal business writing uses many of the same patterns that AI models learned from. If your writing style naturally includes "furthermore", "it is important to note" or "multifaceted", the detector may flag it. High scores do not prove AI authorship — they indicate the presence of patterns common in AI text.

Commercial detectors use fine-tuned language models trained on known AI vs. human text pairs, making them more accurate. This tool uses rule-based heuristics that run entirely in the browser without any API call or server. The tradeoff is privacy and speed versus precision. Use this for a free, instant first pass; use a commercial tool when accuracy is critical.