Artificial intelligence (AI) is increasingly deployed in psychological counseling, offering low-cost, anonymous, and on-demand support for individuals who face barriers to traditional mental health services. Acceptance of such systems, however, depends on more than the availability of AI tools. The Technology Acceptance Model and related frameworks have emphasized cognitive appraisals, perceived usefulness, and relatively stable attitudes toward technology, but have paid less attention to users' psychological state at the point of service contact. This gap is particularly important in mental health contexts, because many prospective users seek counseling precisely when they are distressed and in need of emotional support. Drawing on social baseline theory and dual-process theory, this study examined whether psychological distress directly reduces users' immediate evaluations of an AI psychological counselor and whether it weakens the process by which pre-existing attitudes toward AI and psychological counseling translate into such evaluations.

This study conducted an integrative secondary analysis of two independently collected datasets, yielding a total sample of 490 participants. Participants first reported demographic information, subjective socioeconomic status, attitudes toward AI and psychological counseling, and current psychological distress using validated measures. They then read dialogue materials attributed to an AI psychological counselor, imagined interacting with the counselor, and immediately evaluated it on five dimensions: likability, perceived empathy, willingness to trust, perceived authenticity, and willingness to use. In Study 1, participants completed four rounds of this imagined interaction procedure with different AI counselor materials, and average ratings were used in the present analysis; in Study 2, participants completed the procedure once. Analyses of covariance and linear regression models tested the direct effect of distress, the predictive effects of pre-existing attitudes, and the moderating role of distress, controlling for data source, gender, age, education, and subjective socioeconomic status.

Psychological distress showed a selective but meaningful negative association with immediate evaluations of the AI psychological counselor. In continuous regression analyses, higher distress significantly predicted lower likability, lower perceived authenticity, and lower willingness to use, whereas its direct effects on perceived empathy and willingness to trust were not significant. Group comparisons based on clinical distress categories showed that participants with clinically meaningful distress rated the AI counselor lower on likability, perceived empathy, and willingness to use than those without distress, although mild, moderate, and severe distress groups did not differ significantly on these indicators. Perceived authenticity showed a clearer graded pattern, declining as the distress level increased. Attitudes toward AI positively predicted all immediate evaluation indicators, while attitudes toward psychological counseling positively predicted likability, willingness to use, and perceived empathy. Psychological distress further moderated the attitude--evaluation link: as distress increased, the positive associations between both AI and counseling attitudes and likability, willingness to trust, willingness to use, and perceived empathy weakened. This moderating effect was not significant for perceived authenticity.

These findings indicate that acceptance of AI-based psychological counseling cannot be adequately predicted by general attitudes toward technology or counseling alone. Users' psychological distress at the moment of engagement has independent explanatory value and may alter the process through which pre-existing attitudes are converted into immediate experience. Consistent with the Social Baseline Theory, elevated distress may intensify users' need for genuine interpersonal connection, making an artificial counselor less able to satisfy relational expectations. From a dual-process perspective, distress may also consume cognitive resources and weaken reflective processing, allowing immediate affective reactions to play a larger role in evaluation. The exception observed for perceived authenticity suggests that authenticity judgments may involve a relatively distinct process, possibly because users evaluate whether an AI counselor's responses align with expectations for a nonhuman service provider rather than relying solely on general attitudes. Overall, the results identify psychological distress as a boundary condition for applying traditional technology acceptance models to AI mental health services.

Practically, this study suggests that individuals with higher psychological distress may find it more difficult to trust and like an AI psychological counselor, even when they hold positive prior attitudes toward AI and psychological counseling. Therefore, the promotion and application of AI-based mental health services should not follow a one-size-fits-all approach. Service procedures should first consider users' level of psychological distress. For users without evident distress, AI psychological counselors may serve as convenient preventive tools. For users with elevated distress, however, it may be necessary to introduce human-AI collaboration or professional referral mechanisms in a timely manner, so that the safety and effectiveness of mental health services can be better ensured.