WordBinary

WordBinary Learning Centre

AI Detection Guides: Scores, Reports, Accuracy & False Positives

Understand AI detection scores, sentence-level signals, model testing, multilingual analysis, false positives and responsible report interpretation.

Start Here

AI detection estimates whether patterns in a piece of writing resemble patterns learned from AI-generated and human-written text. It does not identify an author, prove that a particular model was used or establish academic misconduct. The most useful starting point is therefore to understand what the detector measures and what evidence remains outside the score.

An AI probability is a model output, not the literal percentage of words written by AI. Sentence-level signals can help a reviewer locate passages contributing to a document result, but those passages still need to be considered with drafts, sources, version history, language background and the applicable institutional or editorial policy.

False positives matter because formal, translated, formulaic or highly regular human writing can sometimes resemble generated prose. Edited or mixed human-AI writing can also produce different results across tools. WordBinary reports are designed to support contextual review rather than replace authorised human judgement.

Use the pathways below to learn how detection works, interpret a score, investigate a possible false positive or review a report. Dated WordBinary evaluations are provided separately so that model findings remain connected to their datasets, measures and stated limitations.

Understanding AI Detection

AI Scores and Reports

False Positives and Limitations

Evidence from WordBinary Research

Published AI Detection Evaluations

Read the methodology, dataset, measures, results and limitations behind WordBinary's published model evidence.

Continue with WordBinary

AI DetectorAI detection researchAll resources