Reference Glossary

AI Detection Glossary

Every important concept in AI detection, AI writing analysis, and AI-generated text evaluation, defined clearly. Built for educators, students, and anyone who needs to understand how detection systems work.

Terms defined

Perplexity

measures how predictable a piece of text is to a language model. AI-generated text often has lower perplexity because language models are trained to output the most statistically probable word sequences.

Read definition

Detection Concepts

AI Watermarking

is a technique that embeds hidden signals into AI-generated text at the point of generation, allowing the text to be reliably identified as AI-produced even after editing or paraphrasing.

Read definition

Detection Concepts

False Positive

occurs when a detector incorrectly flags human-written text as AI-generated. It is one of the most consequential error types in academic settings, where an incorrect flag can have serious consequences for students.

Read definition

AI Writing Concepts

Hallucination

refers to the tendency of AI language models to generate factually incorrect or entirely fabricated information presented with apparent confidence. In academic contexts, hallucinated content poses serious risks to research quality and citation accuracy.

Read definition

Browse by category

Academic Integrity

3 terms

Academic Integrity Contract Cheating Plagiarism

AI Models & Tools

2 terms

ChatGPT GPT-5

AI Writing Concepts

5 terms

Fine-Tuning Hallucination Large Language Model (LLM) Prompt Engineering Training Data

Detection Concepts

7 terms

AI Watermarking Confidence Score Detection Accuracy False Negative False Positive Sentence-Level Detection Stylometry

Detection Signals

3 terms

Burstiness Perplexity Token Probability

All terms A–Z

A 2 terms

Academic Integrity is the commitment to honesty, fairness, and ethical conduct in all academic work. In the context of AI, it encompasses the policies and practices that define acceptable and unacceptable uses of AI tools in educational and research settings. Academic Integrity AI Watermarking is a technique that embeds hidden signals into AI-generated text at the point of generation, allowing the text to be reliably identified as AI-produced even after editing or paraphrasing. Detection Concepts

B 1 term

Burstiness measures the variation in sentence complexity and length across a piece of text. Human writing naturally alternates between short, punchy sentences and longer, more complex ones - AI-generated text tends to maintain a uniform rhythm throughout. Detection Signals

C 3 terms

ChatGPT is a conversational AI assistant developed by OpenAI, built on large language models including GPT-4 and GPT-5. It is the most widely used AI writing tool among students and one of the primary targets of AI detection systems in academic settings. AI Models & Tools Confidence Score is a numerical value indicating how likely a piece of text is to have been AI-generated. Typically expressed as a percentage, it represents the detector's degree of certainty based on the statistical signals it has measured. Detection Concepts Contract Cheating occurs when a student arranges for another person - or, increasingly, an AI system - to complete academic work on their behalf and submits that work as their own. It is considered one of the most serious forms of academic misconduct. Academic Integrity

D 1 term

Detection Accuracy refers to how reliably an AI detection tool correctly identifies AI-generated text while avoiding false positives on human-written text. It is typically expressed as a combination of sensitivity (catching AI text) and specificity (not flagging human text). Detection Concepts

F 3 terms

False Negative in AI detection occurs when a detector fails to identify text that was actually AI-generated. The text passes as human-written despite being AI-produced, fully or in part. Detection Concepts False Positive occurs when a detector incorrectly flags human-written text as AI-generated. It is one of the most consequential error types in academic settings, where an incorrect flag can have serious consequences for students. Detection Concepts Fine-Tuning is the process of further training a pre-trained large language model on a specific dataset to improve its performance on a particular task, style, or domain. Fine-tuned models can produce outputs that differ significantly in statistical profile from the base model. AI Writing Concepts

G 1 term

GPT-5 is a large language model developed by OpenAI, released in August 2025, and the current engine powering ChatGPT. It represents a major capability leap over GPT-4 and is now the most widely used AI writing model in academic settings - and the primary target of AI detection systems in education. AI Models & Tools

H 1 term

Hallucination refers to the tendency of AI language models to generate factually incorrect or entirely fabricated information presented with apparent confidence. In academic contexts, hallucinated content poses serious risks to research quality and citation accuracy. AI Writing Concepts

L 1 term

Large Language Model (LLM) is an AI system trained on vast quantities of text to generate, summarize, translate, and respond to language. LLMs are the underlying technology behind tools like ChatGPT, Claude, Gemini, and Copilot. AI Writing Concepts

P 3 terms

Perplexity measures how predictable a piece of text is to a language model. AI-generated text often has lower perplexity because language models are trained to output the most statistically probable word sequences. Detection Signals Plagiarism is the act of presenting someone else's words, ideas, or work as one's own without proper attribution. In the AI era, it has expanded to include AI-generated content submitted as original student work, creating new challenges for academic integrity enforcement. Academic Integrity Prompt Engineering is the practice of crafting inputs to AI language models to produce more accurate, specific, or useful outputs. In academic contexts, it is both a legitimate skill and a mechanism for generating more convincing AI-written content that may be harder to detect. AI Writing Concepts

S 2 terms

Sentence-Level Detection assigns AI probability scores to individual sentences rather than producing a single score for the entire document. This approach identifies exactly which parts of a text are likely AI-generated, making it far more useful for educators reviewing mixed or partially AI-assiste Detection Concepts Stylometry is the statistical analysis of writing style used to identify authorship patterns based on vocabulary, sentence structure, punctuation habits, and other measurable features of an individual's writing. In AI detection, it helps distinguish AI-generated text from a specific writer's establi Detection Concepts

T 2 terms

Token Probability measures how likely each word or word fragment in a piece of text is, given what came before it. AI-generated text tends to have high average token probability because language models are trained to select the most statistically likely next word at each step. Detection Signals Training Data is the large corpus of text used to train large language models. The content, quality, and diversity of training data directly shape a model's capabilities, knowledge, and writing style - and explain why AI-generated text has the statistical properties that detection systems measure. AI Writing Concepts

FAQ

Common questions about AI detection

New to the field? These questions cover the fundamentals of how AI detectors work and what their outputs mean.

What is AI detection?

AI detection is the process of analyzing a piece of text to determine whether it was written by a human or generated by an AI language model such as ChatGPT, Claude, or Gemini. Detectors use statistical signals like perplexity and burstiness to make this determination.

How do AI detectors analyze text?

AI detectors compare the statistical properties of a text sample against what a language model would predict. Key signals include token probability, sentence-level perplexity, and burstiness. Some detectors also use stylometric features and classifier models trained on known AI and human writing samples.

Why do AI detectors sometimes flag human writing?

This is called a false positive. It happens when a human writer naturally uses predictable, formulaic language that shares statistical properties with AI-generated text. Non-native English speakers, technical writers, and students who follow strict essay formats are especially vulnerable to false flags.

Is AI detection 100% accurate?

No AI detector achieves 100% accuracy. All detectors produce some rate of false positives and false negatives. Detection results should be treated as probabilistic indicators, not definitive proof, especially in high-stakes settings like academic discipline cases.

What is perplexity and why does it matter?

Perplexity measures how surprising a sequence of text is to a language model. AI-generated text tends to have low perplexity because AI systems are trained to produce the most probable outputs. This makes perplexity one of the most reliable statistical signals used in detection systems.

Proofademic

Curious how AI detectors evaluate your writing?

Test any text with Proofademic's AI detection tools. Get sentence-level analysis, detection scores, and detailed breakdowns — so you understand exactly what detectors are seeing.

Try Proofademic Free