Abstract: Methods, apparatuses, and systems for return-link optimization are provided. Embodiments identify upload-after-download content (e.g., file sharing content) upon download, and generate one or more identifiers characterizing the content (e.g., a digest). The identifiers are stored in a client-side server dictionary model reflecting a presumption that the content is stored in a server-side dictionary. When content is later uploaded, the server dictionary model is used to identify when the upload content matches previously downloaded content. When a match is detected, the stored identifiers are used to generate a highly compressed version of the upload content, which is then uploaded to the server instead of uploading the full content data. In some embodiments, similar techniques are used to optimize return link bandwidth usage for upload-after-upload transactions.