This research examines what influences the effectiveness of current AI text detection methods in security and education contexts.

• Model size and recency significantly impact detectability - newer and larger models produce text that's harder to identify as AI-generated
• Prompt engineering techniques like role-playing and few-shot prompting can help AI text evade detection
• Post-processing methods including paraphrasing and human editing further decrease detectability
• Current detection methods face significant limitations when confronted with these evasion techniques

Why it matters: As AI-generated content becomes increasingly sophisticated, better detection systems are crucial for combating fraud, academic dishonesty, and misinformation campaigns.

Detecting AI-Generated Text: Factors Influencing Detectability with Current Methods