Different functions have been proposed for the hand gestures speakers spontaneously produce while speaking. The Information Packaging Hypothesis (Kita, 2000) states that gestures can structure rich spatio-motoric information into packages suitable for speaking. It therefore predicts that how information is divided over different gestures affects how it is divided over different processing units in speech: clauses. We indeed found that if participants were asked to express the manner and path of a motion in one gesture, they were also more likely to conflate this information into one clause in speech, whereas if they were asked to produce separate gestures, they were more likely to express manner and path in separate clauses too. These results support the view that there are speaker-internal motivations for gesture production. They confirm predictions made by the Information Packaging Hypothesis, which the Lexical Retrieval Hypothesis and the Image Activation Hypothesis do not make.