Misattributed quotes, including deepfakes (doctored videos) are quickly becoming a dangerous political weapon. We propose and test a new way to deal with them: attribution prompts. By analogy to accuracy prompts proposed by Pennycook and colleagues, we ask the audience whether they believe the alleged source has really made this statement (an attribution question). In a series of pre-registered online experimental studies featuring various quotes about the Covid pandemic, we confirm the hypothesised effects of attribution prompts. Compared to the control condition with no attribution question, we find the following effects. When quotes are negative (signalling that speaker lacks competence or empathy) and misattributed, attribution prompts reduce willingness to share the quote. They also improve speaker’s reputation (by reducing the harm caused by the incorrect attribution of the negative statement). When quotes are positive (signalling competence and/or empathy) and misattributed, attribution prompts still reduce willingness to share but worsen speaker’s reputation (by reducing the improvement caused by the incorrect attribution of the positive quote). When quotes are positive and correctly attributed, there is limited impact on sharing and no impact on reputation.