X Platform Launches Two Timeline Recommendation Adjustments
X Platform has launched two adjustments to its timeline recommendations. The new AI and super-intelligent sorting models can more accurately predict the time users may spend, and the retrieval model is trained to preferentially push posts that users are likely to read in full.
Elon Musk posted that the new sorting model is optimized for "probable time spent," and the retrieval model is specifically trained to surface content that users will read completely. These two changes are driving exponential growth in User Active Seconds.
Charts show that the daily active seconds curve for 2026 is significantly higher than the same period in 2023 to 2025, with a spike in March followed by an overall upward trend. After entering September, it has continued to rise, approaching or exceeding 500 billion seconds by late October, with daily records being set almost every day.
In previous years, daily active seconds fluctuated between 320 billion to 400 billion seconds. In 2026, most periods have stabilized above 400 billion seconds, recently accelerating past previous highs, forming a clear structural uplift.
The platform previously directed its recommendation system goals towards "unregretted user seconds." This update places the prediction of stay duration and completion reading rate as more core signals, with the retrieval layer prioritizing content that can be fully read rather than just high-interaction short posts.
Funding for advertisers and creators is being redistributed: creators willing to invest in long-form and in-depth content are receiving more exposure, while the distribution weight of short, quick-interaction content is relatively decreasing. The platform's advertising inventory is expanding due to the increase in total stay duration, but advertisers relying solely on clicks are under pressure.
Source: Public Information
ABAB AI Insight
Since Elon Musk's acquisition in 2022, he has continuously replaced traditional DAU with "unregretted user minutes/seconds" as the core optimization metric. In 2023, he publicly changed the calculation of user seconds to be based on iOS/Android screen time reporting to reduce bot interference. In 2024, he made multiple announcements about daily active seconds exceeding 400 billion and pushed for the recommendation system to transition from heuristic rules to pure AI.
This adjustment of sorting and retrieval models essentially shifts computational resources further towards the Grok-related prediction and retrieval stack, directing traffic towards content that can be fully consumed through more accurate stay duration predictions, thereby increasing total active seconds, a metric that directly supports advertising inventory and data flywheel, while paving the way for the future complete opening of user-customized feeds (through conversational adjustments).
A similar path can be seen in TikTok's early extreme optimization of completion rates and stay durations, as well as YouTube's emphasis on watch time rather than just clicks. Currently, X is in an expansion phase from "interaction-driven" to "deep consumption-driven," attempting to re-establish platform control between information density and attention duration.
This represents a shift in pricing power: the recommendation system is moving from "who interacts more" to "who can really get users to finish watching," transferring distribution rights from high-interaction accounts and advertising algorithms to content providers that can stably generate predictable stay durations. Mechanically, it replaces manual rules and short-term interaction signals with AI prediction accuracy, shifting the platform's pricing ability for attention towards long-term consumption value.
ABAB News · Cognitive Laws
Stay duration prices content; interaction is just noise.
Predicting time is more valuable than calculating clicks.
Exponential growth comes from models, not users.