We sent 23 AI agents to our website. Most looked human (privatrak.com)

A recent experiment revealed that most AI agents browsing the web appear indistinguishable from human visitors. By testing 23 different AI agents against a custom website, the team aimed to understand how web analytics perceive these visits. Most agents, including those from ChatGPT Work and Claude, reported normal Chrome behavior, avoided signaling automation, and originated from familiar home IP addresses, rendering them effectively invisible to standard bot detection practices. Only a few, like Manus and Claude Cowork, indicated their automated nature, with Manus uniquely signing its requests for verification. This finding is significant for the AI and machine learning community as it highlights the challenges of distinguishing between human and AI-generated traffic, raising ethical questions around web analytics and data privacy. The implications for businesses that rely on accurate user metrics are profound; as AI agents increasingly mimic human behavior, traditional methods of filtering out bots may no longer suffice. The study emphasizes the need for more robust detection techniques and transparency standards in AI interactions online, especially as AI technology continues to evolve and integrate into everyday web tasks.