Abstract: Techniques for improving privacy protection by requesting consent to review and analyze a natural language interaction. A natural language processing (NLP) system may process a user query, provide a system response, and then request feedback indicating whether the system response was responsive to the user query. To improve an accuracy of the NLP system, the NLP system may request consent to review interaction data and generate training data in order to train a model. If consent is granted, the NLP system may store confirmation of the consent and share the interaction data for training. The NLP system may process the user query using a first skill but request the feedback using a separate feedback skill. Based on the feedback, the feedback skill may pass the interaction back to the first skill differently and/or cause the first skill to perform one or more different actions.