Unique fingerprints AI systems in their writing: how to recognize and improve

Researchers have found that different AI systems create text with unique, identified templates. Analyzing these templates, they reached 97.1% of accuracy in determining that AI wrote a specific piece of content.

This is for two reasons: for readers - since web is increasingly filled with AI -generated content, knowledge of how to detect it, helps you evaluate the sources of information. For writers - understanding these templates can help you better edit AI -generated drafts so that they sound more human and authentic.

🚀 Every large AI system has specific written habits that give it out. Researchers have found that these templates remain even in the rewritten content: "These templates are stored even when the texts are rewritten, translated or generalized by external LLM, which indicates that they are also encoded in semantic content."

• 📌 Chatgpt: often uses transitional words such as "definitely", "such as" and "in general."
• 📌 Claude: Uses language as "according to text", "based" or "here's a resume".
• 📌 Grok: can use words as "remember", "possibly", "but also" or "helps in".
• 📌 Gemini: known by using "below", "example", "for example", sometimes in combination with "in resume".
• 📌 Deeseek: uses words as "important", "key improvements", "here's breakdown", "mainly", "etc.".