While the number of services using information technology has increased, artificial intelligence (AI) systems and services are not perfect for certain intelligent tasks, like humans are not perfect sometimes. Mistakes in technology services are frustrating and lead to service cancellations. Additionally, it would be beneficial for system developers to have an effective way for users to tolerate system failures. Therefore, in this paper, we argue that prior information can influence users’ tolerance of such failures. Our hypothesis is that when people are made aware of potential weaknesses in an artificial intelligence system, they are more likely to tolerate an incorrect response if it can be attributed to those known limitations. To verify this hypothesis, we conducted an experiment using a hypothetical webpage. The webpage displayed the “feelings” (emotional responses) that many users would normally describe having when looking at an image. In the experiment, the webpage simulated image recognition of user-uploaded images, while the accuracy of the responses was strictly controlled. Three ways of providing prior information were prepared: Type 1: No explanation; Type 2: The system is not yet perfect; and Type 3: The system does not perfectly identify hidden objects. The results demonstrated a different distribution between the users’ evaluations for Type 3 from Types 1 and 2, which exhibited a typical distribution. Additionally, providing detailed information about the system’s limitations was shown to mitigate system failure. The proposed approach may be an effective way for imperfect systems to induce certain attributions and avoid system evaluation reduction.