Daily health self-management, such as the harmonization of food, exercise and medication, is a major problem for a large group of older adults with obesity or diabetics. Computer-based personal assistance can help to behave healthy by persuading and guiding older adults. For effective persuasion, the assistant should express social behaviors (e.g., turn taking, emotional expressions) to be trustworthy and show empathy. From the motivational interviewing method and synthetic assistants' literature, we derived a set of social behaviors, and implemented a subset in a physical character, a virtual character and a text interface. The first behavior type concerns conversing with high-level dialogue (semantics, intentions), which could be implemented in all 3 assistants. The other behavior types could only be implemented in the characters: showing natural cues (e.g., gaze, posture), expressing emotions (e.g., compassionate face), and accommodating social conversations (e.g., turn taking). In an experiment, 24 older adults (45-65) interacted with the text interface and one of the characters, conform a "one-week diabetics scenario". They experienced the virtual and physical character as more empathic and trustworthy than the text-based assistant, and expressed more conversational behavior with the characters. However, it seems that the preference of interacting with the character or the text interface was influenced by the conscientiousness of the participant; more conscientious people liked the text interface better. Older adults responded more negative to the characters that lacked the social behaviors than to the text interface. Some differences between the virtual and physical character probably occurred due to the specific constraints of the physical character. © 2009 Elsevier Ltd. All rights reserved.