How We Check AI Writing
A practical explanation of our detection workflow, the two key result indicators, and the limits of interpreting AI-generated-text signals.
The Detection Process
Our checker is designed to analyze text supplied through the form on our homepage. It does not claim to connect to a school’s Turnitin account or access proprietary Turnitin reports.
- Text submission. A user pastes an eligible prose passage and starts a check.
- Independent model analysis. The text is sent for analysis by a third-party AI detection provider through a server-side integration when the service is operational.
- Results. The model may return a document-level estimate, a share of text in flagged sentences, and sentence-level indicators.
- Human review. Users are encouraged to interpret highlighted passages in context rather than treat the numerical result as conclusive proof.
Understanding the Two Indicators
Overall AI Likelihood
The overall score estimates how strongly the text as a whole resembles AI-generated writing according to the model. It does not indicate the exact percentage of words written by AI and is not an official Turnitin score.
Text in Flagged Sentences
This metric indicates the share of non-whitespace characters in sentences that the model marks at or above its sentence-level threshold. A mixed-authorship draft may have a low overall score and still include individual sentences that the model flags. The two indicators answer different questions and should be read together.
Sentence-Level Highlights
Where sentence scores are available, highlighted passages show which parts of the submitted text triggered the model’s flagging threshold. A highlighted sentence is a reason to review the language, not proof of its origin.
Why Scores Can Differ
AI detectors use different models, training material, scoring rules, and update schedules. They may also differ in how they handle quotations, tables, short text, academic phrasing, or mixed writing styles. For that reason, a score from this website should not be equated with any institutional Turnitin result.
Limitations and False Positives
All AI writing detectors are susceptible to false positives and false negatives. Conventional academic language or formulaic phrasing may be flagged even when written by a human. AI-generated passages may also escape detection, especially after substantial revision or when little context is available.
How to use the result: Check the surrounding paragraph, examine your reasoning and citations, and refer to your drafts or revision history where useful. Do not make academic misconduct accusations based only on a detector score.
Privacy and Data Handling
Text entered into the checker may be transmitted to a third-party detection provider for processing. Users should not submit confidential or personal information without authorization. Specific processing, retention, and security details will be described in our Privacy Policy once the live integration is finalized.
What We Do Not Test or Guarantee
This page describes our detection workflow, not a proprietary accuracy benchmark. We do not publish numerical accuracy or agreement claims without reproducible labeled evaluations. We do not guarantee parity with Turnitin, identification of the generating AI model, or a particular outcome for an academic submission.