AI Detection Glossary

Perplexity describes how predictable a sequence of words is to a language model. It is one signal discussed in AI detection, not proof that a person or model wrote a passage.

Burstiness describes variation in sentence length and structure across a passage. It can help explain why writing feels uniform or varied, but it cannot establish authorship on its own.

A false positive occurs when human-written text is incorrectly classified as AI-generated. Detector results should be reviewed with drafts, sources, and context.

A similarity score estimates how much submitted text overlaps with indexed sources. It is different from an AI-writing indicator and should be interpreted separately.

A Turnitin AI score is a probabilistic indicator estimating whether qualifying prose resembles AI-generated writing. It is not definitive proof of misconduct.

LLM watermarking refers to statistical signals designed to make generated text easier to identify. Availability, reliability, and detection behavior vary by system.

AI humanization describes revising AI-assisted text to improve clarity, rhythm, and natural expression. Any revision still requires human review and policy compliance.

AI paraphrasing changes wording or structure while preserving a general meaning. It can introduce errors, alter citations, or retain predictable patterns.

An AI detector analyzes text for patterns associated with generated writing. Results are estimates and should be combined with human review and writing-process evidence.