Explore English writing patterns with an experimental, browser-only AI-style score. This heuristic is not an AI probability or proof of authorship, and short-text results are reference only.
Your content stays on your device Free, with no account
Explore writing patterns Experimental · English prose
STYLE SIGNALS, NOT PROOF
A style check, with limits.
This tool scores linguistic patterns. It cannot establish whether a person or an AI wrote a passage.
EXPERIMENTAL HEURISTIC
AI-style score
0 / 100
Not an AI probability or confidence estimate. Higher scores reflect more of the package’s weighted style patterns; they do not establish authorship.
Short text: under 100 words. This result is for reference only.
Words · approximate
0
Sentences · approximate
0
Local n-gram predictability
0
Sentence-length variation
0
The predictability metric uses patterns within your text. It is not perplexity from a pretrained language model. The variation metric is normalized and can be negative. These metrics are descriptive, not evidence of misconduct.
Reports contain scores and limitations, without your original text.
Read every result with caution
This tool provides reference results based on stylistic analysis and is not a definitive proof of academic misconduct. Results for short texts (under 100 words) are for reference only.
No validated accuracy rate is claimed. Formal writing, edited text, non-native English and different genres may receive misleading scores. Do not use a score as the sole basis for grading, disciplinary or hiring decisions.
Engine: ai-text-detector 1.0.8 · MIT license · self-hosted heuristic library · no model download
A FEW SIMPLE STEPS
How to use AI Text Detector
Paste English prose with at least 50 characters and 10 words. Longer passages still do not guarantee reliability.
Select Analyze text. The self-hosted heuristic library runs in a local worker; you can cancel or edit the text to stop it.
Review the AI-style score and its limitations. A score is not a probability or an authorship verdict. Text under 100 words receives a prominent reference-only warning.
Copy or download a report containing scores and limitations, without your original passage. Clear removes the text and result.
PRIVATE BY DESIGN
Why use this tool?
Your text is never uploaded to any server; all analysis is completed locally in your browser. The heuristic library is self-hosted, needs no account or model download, and does not save your text in browser storage. Its style score is a reference signal, not proof of authorship.
Tool inputs stay in your browser. Only your theme preference is saved on this device.
A few things worth knowing
Is the score an AI probability or confidence level?
No. The 0–100 value is the ai-text-detector library’s raw weighted style score scaled for display. It is not calibrated to the chance that AI wrote a passage. We do not display the package’s binary AI/human verdict or its confidence label, and no validated accuracy rate is claimed.
Can I use it to prove academic misconduct?
This tool provides reference results based on stylistic analysis and is not a definitive proof of academic misconduct. Results for short texts (under 100 words) are for reference only. A score must not be the sole basis for grading, disciplinary or hiring decisions.
Which engine is used?
This version self-hosts the MIT-licensed ai-text-detector 1.0.8 package. It combines linguistic rules and text statistics rather than a pretrained neural classifier. It does not use Earlybird or another downloaded model. Its local n-gram predictability metric is not language-model perplexity.
Which texts are supported?
It is intended for English prose. The input checks a minimum of 50 characters, 10 English-like word tokens and mostly Latin-script letters, but this is not language identification. Code, lists and other languages are outside its intended use. Formal prose, translations, non-native English and different genres may get misleading scores.
Why are short texts flagged?
Brief passages provide few style observations and are easy to mislabel. Every result under 100 approximate English words is marked reference only. Reaching 100 words does not make a result reliable or establish authorship.
What are the limits and privacy rules?
The tool accepts up to 20,000 UTF-16 text units and 3,000 approximate English word tokens. Inputs are never truncated; shorten oversized passages yourself. Analysis runs in a local worker, with cancellation and a time limit. Reports omit the passage but contain the limitations. Hosting receives ordinary page requests; tool text is not transmitted.
