Plain-English definitions of the terms you will run into across AI detection tools, humanizer products, and research on both. The space moves fast, so this glossary covers the core vocabulary, and the learn-more links take you deeper on each concept.
AI detector
A tool that estimates the probability a given text was generated by a large language model. Most detectors score statistical signals such as perplexity (word predictability) and burstiness (sentence-length variance) and combine them into a score. Leap's detector runs free in your browser, scores text 0-100 from writing signals, and highlights the sentences that weigh most. It is a signal, not proof, and it should never be the only basis for accusing anyone.
AI humanizer
A tool that rewrites AI-generated text so it reads more like natural human writing. Leap's humanizer is a cleanup pass that runs free in your browser: it replaces stock AI phrases, removes em dashes and invisible characters, adds contractions, and marks sentences of uniform length so you can vary them. It does not guarantee passing any detector and does not promise to bypass Turnitin or any other checker.
Burstiness
The variance in sentence length across a passage of text. Human writing is bursty: it mixes short, medium, and long sentences. AI writing tends to be uniform. Detectors measure burstiness as one of the core signals alongside perplexity, and Leap's AI score includes it in the 0-100 scoring.
Copyleaks
An enterprise AI detector and plagiarism checker used by universities and corporate customers. It is a paid product positioned for institutional content verification.
Detection threshold
The score above which a detector labels text as likely AI-generated. Thresholds vary between products and institutions, and text that lands near a threshold is ambiguous, which is why detector output works best as a starting point for human review rather than a final verdict.
Em-dash fingerprint
A well-documented pattern where AI models overuse em dashes for punctuation. It is not a detection signal on its own, but a strong visual tell for human readers. Leap's AI score factors em-dash density into its signals, and the humanizer removes em dashes as part of the cleanup pass.
GPTZero
One of the earliest widely used AI detectors, launched in 2023. It scores perplexity and burstiness and is commonly used by educators.
Hedging
A writing pattern where the author softens claims with qualifiers like often, typically, arguably, or in many cases. AI models hedge heavily to avoid strong claims. Excessive hedging is one of the writing signals Leap's AI score looks at, alongside transition words and stock phrases.
Large language model
A neural network trained on massive text corpora to predict the next token in a sequence. GPT-4, Claude, Gemini, and Llama are all LLMs. Their training objective, predicting the most likely next word, is exactly what produces the uniform statistical patterns detectors score.
Originality.ai
An AI detector and plagiarism checker used by SEO agencies and publishers. It is a paid product marketed for content-team QA and freelancer screening.
Paraphrasing
Rewording text while preserving meaning. Commonly confused with humanization, but paraphrasing alone usually does not change how detectors score a passage, because the underlying statistical patterns, like uniform sentence length and predictable word choice, survive the rewording.
Perplexity
A measure of how surprised a language model is by the next word in a sequence. Low perplexity means the word was expected; high perplexity means it was not. AI text tends to have low perplexity because AI picks expected words, while human text is less predictable.
Plagiarism
Presenting another author's words or ideas as your own. Distinct from AI detection: plagiarism checkers compare text against published corpora, while AI detectors score statistical patterns in the writing itself. Both can flag the same submission for different reasons.
Stylometry
The statistical study of writing style. It uses features like word frequency, sentence structure, and punctuation to characterize authorship. Modern AI detectors are a specialized subset of stylometry focused on distinguishing AI-generated text from human writing.
Token probability
The model-assigned likelihood of a particular word appearing next in a sequence. AI-generated text tends to follow high-probability word sequences, which is why detectors can use token predictability as a signal of machine authorship.
Tier-1 AI vocabulary
The set of high-register words AI models overuse: leverage, navigate, ensure, multifaceted, paramount, facilitate, robust, intricate, delve. Humans use these words occasionally; AI uses them reflexively. Replacing stock AI phrases like these is part of what Leap's humanizer does in its cleanup pass.
Turnitin
The dominant anti-plagiarism platform in higher education. In 2023 Turnitin added AI detection alongside its plagiarism-similarity module, and it is deployed at many universities worldwide.
Undetectable AI
An AI humanizer product launched in 2023, positioned for students and content marketers. It is one of the best-known humanizers by volume.
Watermarking
A technique where an AI model embeds an imperceptible statistical signal in its output that a watermark-aware detector can recover. Research on LLM watermarking exists, but it is not widely deployed: most production models output unmarked text.
See the mechanics on your own text
Run any text through Leap's free AI score checker to see sentence-level scoring based on burstiness, stock phrases, hedging, em-dash density, and repetition. No signup, nothing leaves your browser. Then try the free humanizer to see stock-phrase replacement, em-dash removal, and contractions applied live.