Open-ended questions in web surveys suffer from substantial item nonresponse. For example in a recent survey on a sensitive topic nonresponse reached up to 56%. Our method automatically classifies different types of item nonresponse, allowing us to ask suitable follow-up questions during a web survey and thus reducing final item nonresponse rates substantially. The classification is available for English, German and Spanish and can also be applied to already collected data. The percentage of successfully converting nonresponse into a useful answer varies across questions but is mostly between 20%-40% with exceptions of up to 74%.