Features are disclosed for determining which content item or items are likely to be requested by a particular user or device, or by a group of users or devices. The determined content items may be obtained independently of a request from the users or devices, and substantially current versions of the content items may be maintained at a server system, such as an intermediary system. Visual representations of the content items may be generated to reduce processing requirements at a user device. When the content items are determined to be likely to be requested by a particular user, a predicted time may also be determined, and the visual representations may be generated such that they are available at the predicted time.