PILOT: Control Surfaces for Authoring Social Media Feeds

Abstract

Personalized social media feeds infer preferences from behavior, leaving people little direct control over what they see. Existing controls range from post-level reactions to rules and natural language, but little is known about how people use them together or how added expressiveness changes effort. We built PILOT, a Bluesky feed-authoring system that turns in-feed actions into explicit preferences, deterministic ranking, and inspectable outcomes. A study with seven participants informed an expanded implementation, which we organized into three nested control surfaces that ten participants compared within subjects. Participants assigned controls to distinct jobs: broad controls set direction, in-post actions refined results, and filters removed content. Richer surfaces did not necessarily feel more effortful, and participants' experiences depended more on whether they could verify and repair outcomes. Our findings show that usable feed control requires not simply more controls, but orchestration across mechanisms that support expression, inspection, repair, and episodic use.