Can’t You Tell I am a Human? A Comparison of Common Text and Image CAPTCHAs Using a Low-Fidelity Methodology

Abstract

Many websites employ CAPTCHAs to limit web robot traffic. Al- though CAPTCHA challenges may improve security they come at the cost of reduced usability and accessibility. This study set out to explore users’ performance with two common types of CAPTCHAs that rely on image and text challenges. The results show that partic- ipants resolved image challenges faster than text challenges. Image challenges were also preferred over text challenges. Overall, each challenge took approximately 10 seconds to solve, and preferences scores were moderate. The error rates were high, but no significant difference was observed. We therefore argue for new methods ca- pable of separating humans from robots that do not rely on user intervention.