<p dir="ltr">The literature suggests that AI companions could play an important role in addressing the mental health crisis. However, there is limited research on the factors that may impact the working alliance with AI companions and how clinically validated and non-validated companions compare in their ability to develop a working alliance across varying demographics. This gap is problematic because the strength of the working alliance and patient demographics are closely related to patient outcomes. Therefore, individuals may be using unproven and potentially unsafe solutions. This study examined factors that may influence the working alliance with AI companions across various demographics through a cross-sectional survey of 253 U.S. adults. Depression and anxiety were screened with the PHQ-2 and GAD-2. Working alliance strength was assessed with the WAI-SR. In addition, three open-ended questions inspired by the Technology Acceptance Model were analyzed thematically. Clinically validated companions developed significantly higher working alliances than non-validated companions. Alliance scores did not vary by demographics or anxiety/depression intensity. Frequently mentioned themes across companion types were ease of interaction, a judgment-free environment, and guidance. Validated companions performed better at providing emotional support, serving as an alternative or supplement to therapy, offering help during difficult times, and assisting with planning and setting goals. Non-validated AI companions were praised more frequently for their affordability. Validated AI companions had significantly lower adoption rates than non-validated companions. The findings suggest that most users of AI mental health companions are at risk of suboptimal outcomes, highlighting the need for clinically validated and affordable AI companions, as well as enhanced regulatory frameworks. This study also proposes a revised version of the WAI-SR adapted for AI companions.</p>