AI Text Detector
Detect whether a text was written by AI using 9 AI signals and 6 human markers (irony, informal tone, sourced quotes). 100% local analysis, supports English, French and Spanish.
Identifying AI-generated text has become a daily challenge for teachers, editors, recruiters, and publishers.
9 AI signals and 6 human markers
AI signals: sentence variation (burstiness), lexical richness, LLM markers (hedging), local entropy, paragraph regularity, logical connectors, abstract vocabulary, tone uniformity, and intro-body-conclusion structure. Human markers: sourced quotes with attribution, parenthetical digressions, informal tone and spontaneous expressions, irony and personal voice, proper noun density, dry ending without moral.
100% local and private
The score (0–100%) is calculated entirely in your browser. No text is ever sent to a server. Language — English, French, or Spanish — is detected automatically.
How to interpret the score
≥ 70% → likely AI · 40–70% → mixed or uncertain · < 40% → likely human. Reliability increases with length: 150 words minimum, 500 words for high reliability.
Paste a text of at least 50 words to start the analysis.
AI Text Detector
Why use an AI text detector?
With the rise of ChatGPT, Gemini, and Claude, distinguishing human-written text from AI-generated content has become a critical challenge for educators, editors, and recruiters.
Our bidirectional approach — 9 AI signals combined with 6 human markers — significantly reduces false positives compared to traditional one-directional detectors.
The analysis is 100% local: no text is ever sent to a server, ensuring complete privacy for your sensitive documents.
Use Cases
- Teacher / Professor
- Check whether student essays and papers were written with the help of generative AI before grading.
- Writer / Content Editor
- Verify the authenticity of articles received from external contributors before publishing on your site or blog.
- Recruiter / HR
- Assess whether cover letters and writing test responses from candidates are authentically written.
- Researcher / Journalist
- Analyze text sources to identify potentially auto-generated content as part of an investigation.
How to use this tool
Paste or type the text to analyze (minimum 50 words). Language is detected automatically among English, French, and Spanish.
Review the overall score (0-100%) and the breakdown of 9 AI signals and 6 human markers. Enable 'Structured text mode' for formal documents.
Interpret the result: 70% or above indicates likely AI text, 40-70% is mixed, and below 40% is likely human. Reliability increases with text length.
Frequently Asked Questions
- How does AI text detection work?
- The tool combines two axes. AI axis: nine statistical signals (sentence variation, lexical richness, LLM markers, entropy, paragraph regularity, logical connectors, abstract vocabulary, tone uniformity, template structure) weighted to produce a score from 0 to 100%. Human axis: six markers (sourced quotes, parenthetical digressions, informal tone, irony, proper nouns, dry ending) reduce the score when typically human traits are detected. This bidirectional approach limits false positives.
- What score indicates AI-generated content?
- A score of 70% or above suggests the text was likely generated by an AI. Between 40% and 70%, the result is mixed or uncertain — the text may be an AI rewrite, a highly structured human text, or an academic paper. Below 40%, the text is likely human-written.
- Can this tool detect text from ChatGPT, Gemini, or Claude?
- Yes. The tool detects text produced by the major language models (ChatGPT, Gemini, Claude, Mistral, etc.) because they share the same statistical tendencies: regular paragraphs, hedging markers, frequent logical connectors, and high abstract vocabulary. However, AI text heavily rewritten by a human may score lower — a known limitation of all heuristic approaches.
- How reliable is the analysis?
- Reliability directly depends on text length. For fewer than 150 words, the result is low-reliability — the tool states this explicitly. Between 150 and 500 words, reliability is moderate. Above 500 words, the analysis is considered reliable. The reliability indicator is displayed at all times.
- Why is my academic paper flagged as AI-generated?
- Academic and high-quality journalistic texts often trigger the structural regularity and logical connector signals, because their style is inherently formal and organized. The result typically falls in the mixed zone (40–60%). Enable 'Structured text mode' for a more appropriate analysis of these formats.
- Is the analyzed text sent to a server?
- No. All analysis is performed in your browser using pure JavaScript. No text, no results, and no data are ever sent to any external server.
- Can this detector be fooled?
- To some extent, yes. AI text rewritten by a human — shorter sentences, concrete vocabulary, tone shifts — will score lower. However, the 6 human markers make detection more robust: genuine human text naturally contains irony, informal tone, sourced quotes, or digressions that AI doesn't reproduce. Conversely, a very formal human writer may be flagged as 'Mixed / Uncertain'.
- Which languages are supported?
- The detector natively supports English, French, and Spanish. Language is detected automatically from frequent words in the text (stopwords). Each language has its own marker sets for all 9 signals, ensuring culturally adapted analysis.
Understanding AI Text Detection
How can you tell if a text was written by AI?
An AI text detector analyzes AI signals (paragraph regularity, hedging markers, logical connectors, burstiness) and human markers (irony, informal tone, sourced quotes, digressions) to assign a score from 0 to 100% indicating the likelihood of artificial origin.
What score does ChatGPT-generated text typically get on an AI detector?
Text generated by ChatGPT typically scores above 70% on statistical AI detectors, primarily due to its remarkably regular paragraph lengths and high density of hedging markers.
Are AI text detectors reliable?
AI text detector reliability increases directly with text length: results are considered high-reliability above 500 words. Below 150 words, results are unreliable regardless of the method used.
What is the most important signal for detecting AI text?
On the AI side, structural regularity of paragraphs is the most discriminating signal (32% weight). On the human side, the presence of irony, informal tone, and sourced quotes is the most reliable counter-signal. The tool combines both axes to limit false positives.
Can an AI text detector work without sending data to a server?
Yes. A client-side AI detector computes all statistical signals directly in the browser using JavaScript, without transmitting any text to external servers, ensuring complete privacy.
Tool created on March 18, 2026