Conversational interviewing, in which interviewers and respondents work together to make sure questions are understood as intended, can help respondents answer more accurately than strictly standardized interviewing. It takes longer and can be costly. Here we measured response accuracy and interview length for three kinds of partially conversational interviewing, which resemble current practice in some organizations. Census Bureau interviewers telephoned paid laboratory respondents, who answered factual questions from ongoing government surveys on the basis of fictional scenarios. Interviewers either (1) read scripted definitions of question concepts when respondents explicitly asked for clarification; (2) used their own words to present official definitions of question concepts when respondents explicitly asked for clarification; (3) presented scripted definitions whenever they deemed it necessary, even if respondents hadn't explicitly requested clarification. For all three partially conversational techniques responses were reliably more accurate than for strictly standardized interviews, and interviews took reliably longer; compared to more fully conversational interviews (in which interviewers could present definitions in their own words whenever they thought this might help) responses were less accurate and interviews shorter. Results suggest that response accuracy improves whenever respondents get help, whether or not they ask for it explicitly; but if it is left only up to respondents to ask for help, they often won't.