Glossary

AI Writing Glossary

The vocabulary of AI writing and AI detection, defined in plain English — so when a detector report says "low perplexity", you know exactly what it means.

AI humanizer

An AI humanizer rewrites machine-generated text so it reads like a person wrote it — by changing sentence structure and rhythm, not just swapping synonyms.

AI detector

An AI detector estimates whether text was written by a human or a language model, using statistical signals like perplexity and burstiness plus trained classifiers.

Perplexity

Perplexity measures how predictable text is to a language model. Low perplexity — predictable word choices — is the classic statistical signature of AI-generated text.

Burstiness

Burstiness measures variation in sentence length and structure. Humans write in bursts; AI models keep a steady rhythm — making low burstiness a machine tell.

Large language model (LLM)

A large language model is a neural network trained on vast text to predict the next word — the technology behind ChatGPT, Claude, Gemini and modern AI writing.

Token

Tokens are the word-fragments language models actually read and write. What they are, why limits are counted in them, and what they have to do with detection.

Temperature

Temperature controls how adventurously a language model picks its next word — low is safe and repetitive, high is varied and risky. It shapes how detectable output is.

AI watermark

AI watermarking hides a statistical signature in generated content at creation time. How it works for text, where it's deployed, and why it's fragile.

Stylometry

Stylometry identifies authors by their measurable writing habits — centuries older than AI detection, and the foundation it quietly stands on.

Paraphrasing

Paraphrasing restates text in different words while keeping the structure — which is exactly why paraphrased AI text still gets flagged by detectors.

Prompt engineering

Prompt engineering is crafting model instructions for better output. What it can do for writing quality — and why it can't make text undetectable on its own.

Hallucination

AI hallucination is when a model states false information fluently — invented citations, fake facts, plausible nonsense. Why it happens and why no humanizer fixes it.

Training data

Training data is the text corpus a language model learns from — and the reason AI writing sounds the way it does, from 'delve' to the five-paragraph template.

Zero-shot detection

Zero-shot detection identifies AI text without training a classifier on examples — using a model's own probability landscape, like DetectGPT's curvature method.

Try it on your own text

Humanize a passage and check its AI-probability score in one place — 3-day free trial, and your text is never stored on our servers.