0-30%
Usually less concerning, but still verify if the context matters. A clean score does not prove human authorship.
Glossary
AI detector glossary: the terms users actually need.
Detector pages often throw around technical words without explaining what they mean. This glossary keeps the definitions practical so users can understand scores, caveats and tool claims.
Core termsFalse positive
SignalsPerplexity
SignalsBurstiness
ReminderScore ≠ proof
Terms
Detector terms in plain English.
| Term | Meaning | Why it matters |
|---|---|---|
| AI score | A tool’s estimate that text looks AI-generated. | It is not proof of who wrote it. |
| False positive | Human writing flagged as AI. | Can harm honest writers if treated as proof. |
| False negative | AI writing missed by a detector. | A clean score does not prove human authorship. |
| Perplexity | How predictable the word choices appear statistically. | AI text can be very predictable, but polished human text can be too. |
| Burstiness | Variation in sentence length and complexity. | Human writing often varies more, but genre and editing matter. |
| Classifier | A model trained to sort text into categories. | It learns patterns, not certainty. |
| Sentence highlighting | Passages the detector thinks drove the result. | More useful than a bare percentage because you can inspect the text. |
| Watermarking | A possible hidden signal in generated text. | Not universal and not the same as public detector scoring. |
Score reading
Do not read a detector score as a verdict.
Use the score to decide what to inspect. Short text, formulaic writing, non-native English, heavy editing and translated drafts can all move the result.
30-70%
Read highlighted passages. Look for repeated sentence shapes, vague claims, generic intros and a rhythm that feels too uniform.
70-100%
Review seriously, but do not punish automatically. Ask for drafts, notes, revision history and context before drawing conclusions.
Related guides
Keep going by intent.
These pages cover the common user paths around detectors, false positives, humanizers and policy.
AI detector false positiveWhat to do when human writing is flagged as AI.
Turnitin AI detectionHow to interpret AI indicators in school workflows.
AI detector for studentsDrafts, notes, revision history and fair use.
AI detector for teachersHow to investigate without treating a score as proof.
AI detector for SEOEditorial review, quality signals and content risk.
AI humanizer for businessEmails, proposals, support replies and brand voice.
Before/after examplesSee what real humanizing changes.
GPTZero alternativesOther detectors and when to use them.
ZeroGPT alternativesFree and paid detector alternatives.
FAQ
Quick answers.
What is the most important detector term?
False positive, because it explains why a detector result can harm honest writers if treated as proof.
Is perplexity enough to detect AI?
No. It is one signal among many, and polished human writing can also be predictable.
What is the practical takeaway?
Read detector scores as signals, inspect highlighted text and use human judgment.