Musk Outlines New Ranking and Retrieval Models for X Timeline

SpaceX CEO Elon Musk outlined two changes to X’s timeline recommendations on Oct. 19 ET (Oct. 20 UTC), including a retrieval model aimed at finding posts readers finish and a ranking model meant to estimate viewing time more accurately.

Retrieval gathers candidate posts before ranking. X’s system draws candidates from accounts a user follows and from accounts outside that group. Ranking then helps order the candidates, while selection and visibility filters also shape what appears in a timeline.

Predicted time spent was already among X’s ranking inputs, alongside expected likes, replies and negative feedback. Musk described the new ranking model as an effort to improve that estimate, not to add viewing time as a new signal.

For crypto publishers, retrieval can affect whether a post from an account outside a user’s follow list enters consideration. The post provided no measured results for either model, data on their effect on crypto-post reach or rollout schedule.