Not available
AI writing detection
AI writing detection is not offered. We are evaluating research detectors internally, and none has evidence strong enough to judge a person's work.
Customer status:Not validated. Switched off.
What a detector measures
Zero-shot detectors such as Fast-DetectGPT and Binoculars ask a language model how predictable each word is. Text that a model finds unusually predictable is called machine-like. That is a property of the text and the model, not a record of who wrote it.
Why detectors make mistakes
Predictability changes with topic, genre, length, editing, translation and the writer's vocabulary. A detector tuned on one kind of text can be wrong on another, and a threshold chosen for English does not carry over to Arabic.
Why human writing gets flagged
Formal, formulaic, heavily edited or plainly worded writing can look predictable. Published research has reported higher false-positive rates for writers using English as an additional language. Short texts are especially unreliable.
Why AI writing can pass unnoticed
Sampling settings, newer models, human revision, paraphrasing and translation all change the statistics a detector relies on. A low score is not evidence that a person wrote the text.
Languages under evaluation
Internal research covers Arabic and English separately, with separate thresholds. Other languages are treated as unsupported.
Why the feature is unavailable
We will only consider offering a result after an independent evaluation shows acceptable false-positive rates in each language and genre, with measured uncertainty and human review. That evidence does not exist yet.
How your documents are handled
Your documents are not sent to any AI detection service and are not used to train or benchmark detectors. Similarity reports keep the line "AI writing signals: disabled", and no AI estimate is added to the similarity figure.
What this is not
It is not a plagiarism verdict, not proof of authorship, and not equivalent to any other vendor's product.