Google Discovers Freshness Fix: SDF Cut Stale Content Complaints 54.9%

User complaints about stale content in Google Discover fell 54.9% over two years after Google deployed an AI filtering system called Supersession-Decay Filtering (SDF), according to newly published Google research. The system has been live for two years — a detail that means publishers and SEOs are already operating under its rules without published guidance until now.

The paper, titled Decomposing Staleness in Recommender Systems: A Dual-Filter Framework for Supersession and Decay, explains how SDF works inside Discover, the personalized feed in the Google Search app that serves hundreds of millions of daily active users and billions monthly.

The researchers wrote: "While downstream ranking systems remain engagement-based and trend-aware, SDF acts as a proactive filtering layer. It prunes fundamentally stale content before it enters the computation-heavy ranking stage, providing a refined candidate set."

What does SDF actually filter?

The system uses two filters, and either one can block a page from ranking. Content that survives both remains eligible for Discover.

• Supersession filter: AI evaluates the relationship between older and newer content to detect when fresher information has made earlier content obsolete, and whether that earlier content remains useful.
• Relevance Decay filter: A second AI model predicts when content will naturally lose usefulness over time, even when nothing newer has replaced it.

Once stale content is pruned, downstream ranking systems — including user engagement signals and what Google calls "trend-aware" systems — rank the remaining candidate set. The filtering also cuts computing costs, because outdated pages never reach the ranking stage.

Why publication dates and engagement metrics fail

Google's researchers rejected the two traditional staleness signals. Publication dates only measure when something was published, not whether it remains relevant. Content can turn stale within days of publication when newer information supersedes it. It can also decay gradually without any replacement. Cutoff dates carry a second risk: useful evergreen content gets blocked from display.

Engagement metrics fail in the opposite direction. The researchers found engagement-based systems are reactive — signals arrive long after users have already judged content stale and stopped clicking.

What should publishers do now?

The two-year deployment window means content and interlinking strategies built on freshness-blind assumptions are already out of date, according to the report's analysis. Publishers need answers to three questions:

• What percentage of content should be evergreen, sustaining long-term informational traffic?
• What percentage has a limited relevance window, and is there an update plan when it goes stale?
• What percentage is breaking news with a short relevance span?

Interlinking deserves particular attention. Many plugins and AI-based interlinking tools rely on semantic relevance alone. SDF adds freshness as a second dimension to that calculation, which affects how long an internal link continues to pass value.

Testing showed SDF reduced stale recommendations compared with traditional methods, delivered modest gains in user engagement and improved content diversity. Publishers who audit their archives against the supersession and decay criteria now hold a concrete map of what Discover will and will not surface.