PromptPrint asks a privacy-relevant question: do short, task-driven prompts reveal enough writing style to identify the person behind them? The paper studies more than 20,000 real prompts from over 1,000 users.
The authors report that lexical patterns outperform semantic encoders for this kind of prompt-based identity signal, suggesting that even brief LLM interactions may carry recognizable behavioral fingerprints.
That matters for AI privacy because prompts are often treated as disposable inputs. If prompt style is linkable across sessions or services, anonymization becomes harder than simply removing names or account IDs.