AI Text Detector
Check text for patterns commonly associated with AI writing tools. Scores word choice, sentence structure, and formatting signals, plus a style-similarity breakdown. Free, browser-only, not a proof-of-authorship tool.
The AI Text Detector scans text for patterns that appear more often in AI-generated writing than in typical human writing: word choice, sentence rhythm, punctuation habits, and formatting quirks common to chatbot output. Paste up to 5,000 words directly, or upload a .txt, .docx, or .pdf file and the tool extracts the text automatically before running the same checks.
The analysis is broken into six signal categories so you can see exactly what triggered the score instead of a single unexplained number: lexical patterns (words and phrases that show up disproportionately in AI output), structural patterns (sentence length uniformity and paragraph rhythm), statistical patterns (repetition and predictability measures like burstiness and vocabulary diversity), punctuation and formatting (em dash overuse, invisible characters, and other formatting habits), platform artifacts (leftover fragments like citation markers that sometimes survive a copy-paste from a chat window), and chatbot leftovers (stock phrases like refusal language or offers to help further). Every flagged pattern is shown with the specific words or phrases that triggered it, not just a category label.
Alongside the main score, the tool shows a Style Similarity breakdown - a set of percentages estimating how closely the writing resembles the typical output patterns of well-known AI models, labelled as stylistic similarity rather than detected authorship. Identifying which specific AI model wrote a text with real confidence requires trained classifiers and access ordinary heuristics do not have, so these percentages are capped well below full confidence and treated as a rough style comparison, not an identification.
This tool is a heuristic scanner, not a proof-of-authorship system, and it should never be the sole basis for an academic, employment, or legal decision. Independent audits of commercial AI detectors have found false-positive rates ranging from roughly 5% to 50% depending on the study, and non-native English writers and some neurodivergent writing styles are flagged more often than average because their natural rhythm and vocabulary choices overlap with the same signals the tool looks for. A high score means the writing shares patterns with AI output, not that a person did or did not write it.
Everything runs locally in your browser. Pasted text and uploaded files are never sent to a server, analyzed by a third party, or stored anywhere - closing the tab clears everything.
Frequently Asked Questions
- How accurate is this AI text detector?
Not accurate enough to be treated as proof, and no AI detector on the market is. Independent studies of commercial tools such as GPTZero, Turnitin, and Originality.ai have found false-positive rates ranging from about 5% up to 50% depending on the text sample, and non-native English writers are flagged at especially high rates - one widely cited study found 61% of TOEFL essays written by non-native speakers were flagged as AI-generated by commercial detectors. This tool uses published heuristics, not a trained classifier, and should be treated as one signal among several, never as the deciding factor in an academic, employment, or legal situation.
- Does a low score mean the text was definitely written by a human?
No. A low score means the text does not match the patterns this tool looks for, but a careful human writer can produce text that scores low on AI-pattern checks, and a light edit of AI-generated text can also lower the score without changing who actually wrote it. Absence of flagged patterns is not proof of human authorship, in the same way flagged patterns are not proof of AI authorship.
- What is the Style Similarity score, and does it identify which AI wrote the text?
No - it estimates stylistic overlap with the typical writing patterns of well-known AI models, labelled as a percentage of stylistic similarity, not as a claim about which model was actually used. Reliably attributing text to a specific AI model requires trained machine learning classifiers with access to model-specific training data, which is beyond what pattern-based heuristics can support. Treat the ranked list as a rough style comparison, capped well below full confidence, not an identification.