AI Writing Detection Glossary: 10 Terms Explained
AI detection terms describe measurements and model outputs, not proof of authorship. A false positive is a human text incorrectly flagged as AI; perplexity measures predictability for a model; and a threshold determines how a score becomes a label. These terms are easier to interpret when the tool explains its definitions and tests.
What do the common error terms mean?
These terms describe how a detector performs against examples whose labels are known. They are not interchangeable, and each depends on the test set and classification threshold.
- False positive: human-written text incorrectly labeled AI-generated.
- False negative: AI-generated text incorrectly labeled human-written.
- True positive: AI-generated text correctly labeled AI-generated.
- True negative: human-written text correctly labeled human-written.
- Confusion matrix: a table of true and false predictions for each class.
- Threshold: the cutoff a model uses to turn a score into a category.
What do the model and language terms mean?
A detector’s technical label describes part of the system that produces its result. Different tools may combine techniques or keep their implementation private, so a term should not be assumed to describe every product.
- Classifier: a model that assigns an input to one or more categories.
- Perplexity: a measure tied to how predictable a sequence is to a language model.
- Burstiness: a broad term for variation or clustering in a signal; writing tools may define it differently.
- Watermark: a detectable signal intentionally embedded during generation, when the generation system supports it.
- Provenance: information about where content came from or how it was created.
How should you use these definitions?
Use the provider’s documentation to understand its score and the test conditions behind any accuracy claim. A term such as “confidence” or “AI percentage” may have a product-specific meaning. For high-stakes decisions, pair automated outputs with process evidence and human review.
Common questions
Does a 90% detector confidence mean a 90% chance the author used AI?
Not necessarily. Score meanings and calibration differ by product; consult the tool’s documentation.
Are perplexity and burstiness standardized detector metrics?
No. The terms have broader statistical meanings, and individual tools may define their use differently.
Sources and further reading
Related Ghostiq tools
- AI writing detector — review an estimate, not proof of authorship.
- Text revision tool — review every suggested change for meaning and accuracy.