Objective: The peer interaction–based online model has been influential in the recent development of diabetes management. This model “extends and innovates” the traditional mode of doctor–patient guidance, transforming it into a mode in which both doctor–patient guidance and patient–patient interaction coexist; this new mode has the added advantage of offering “extended continual intervention.” This study contributes to research on extending diabetes management models by investigating how patients with diabetes or prediabetes interact in online health communities, focusing on the interrelationship between self-efficacy characteristics and online participation during patient–patient interactions. Methods: In this cross-sectional study, participants with diabetes of various severities completed an electronic questionnaire, which was formulated with a revised classical scale. The questionnaire was disseminated through diabetes online health communities. Its content covered the respondent’s general condition, self-evaluation of their self-efficacy, and participation in online health communities, specifically with respect to factors such as the time spent in online information each day, the number of groups joined, and the extent of interaction in diabetes online health communities, etc. The main observation indicators were the participants’ self-efficacy, their extent of online participation, and the characteristics of online health communities. Descriptive statistics, chi-square test, linear trend estimation, and ordinal logistic regression were used to explore the relationship between the three indicators. Results: The self-efficacy scores ( ± s) were 51.9 ± 9.12, and 59.1% of interviewed participants had self-efficacy scores greater than the mean. Overall, most participants (96%) considered online diabetes social platforms to be helpful. Groups differed with respect to interaction mode, which indicated that people with high self-efficacy tend to employ various modes of interaction. Participants with high self-efficacy were also more likely to live in cities (p < 0.05) and be married (p < 0.05) and tended to spend more time paying attention to group information (p < 0.05), spend more time viewing group information (p < 0.05), and have a greater degree of interaction with group members (p < 0.05). Information sources for the different grades of participants was primarily obtained from social media. Conclusion: Among people with diabetes, the frequency and intensity of online interaction might positively affect self-efficacy and, by implication, diabetes self-management. Diabetics with high self-efficacy also tend to have positive online interaction and adopt different ways of interaction. In addition, the diabetes information sources of the respondents mainly come from social networks, most of the respondents think that online social networking sites have a positive impact on diabetes self-management, which shows that social network plays an important role in diabetes information source of diabetics. However, the design of online health communities has room for improvement, specifically with respect to the provision of information that patients require. As an interesting side note, among people with diabetes or prediabetes, those who lived in urban area and were married, those who paid more attention to group information, and those who actively participated in interactions tended to have relatively high self-efficacy. The results suggest that people with diabetes have higher-quality self-care when they engage in online health community interactions; such benefits cannot be obtained from treatment in a hospital. In general, enhanced self-efficacy in people with diabetes enables them to more readily acquire diabetes-related knowledge. Online interaction with diabetics, who has the same experience, can not only get more information, but also have a sense of identity and belonging, which enhances self-efficacy and further urges them to actively participate in online interaction. Therefore, online health communities are an important supplement to the clinical treatment of diabetes mellitus and clinicians can take advantage of the educational function of online diabetes groups in their provision of tailored diabetes interventions and take into account the factors that affect the self-efficacy of diabetics (including the frequency and intensity of online interaction, age, marital status, residential area, etc.), to provide tailored diabetes interventions for diabetics. Such a use of online diabetes groups can strengthen diabetes self-management.