Table of Contents
For two years, Google has been quietly deciding which stories in your Discover feed count as “stale” and filtering the losers out before they ever reach a ranking algorithm. Nobody outside Google knew the system had a name, let alone that it had been running since 2024. That changed this week, when a Google research paper surfaced the mechanics of a system called SDF (Supersession-Decay Filtering), and the SEO world is only now catching up to something that’s already been shaping Discover traffic for two years.
What SDF actually is
SDF stands for Supersession-Decay Filtering. Search Engine Journal’s Roger Montti reported on October 21, 2026, that the details come from a Google research paper titled “Decomposing Staleness in Recommender Systems: A Dual-Filter Framework for Supersession and Decay,” hosted on arXiv. The paper describes SDF as “a unified staleness filtering system deployed in Google Discover,” Google’s personalized feed of open-web content inside the Google Search app.
The system addresses two separate reasons content goes stale, and treats them as two different problems rather than one:
Supersession is what happens when newer information makes an older piece of content outdated. Think of a “best laptops of the year” roundup that’s still technically live and well-linked, but has been quietly overtaken by a more current version, elsewhere, that covers newer models. The first filter evaluates the relationship between older and newer content to judge whether the older piece still holds value for a reader.
Relevance Decay is different. It’s content that loses usefulness over time on its own, independent of whether anything newer has replaced it. A seasonal guide, a pricing page tied to an offer that’s expired, or a how-to built around a tool that’s since changed its interface can all decay in relevance even if no competing article ever supersedes them. The second filter predicts when this kind of decay is likely to happen.
If either filter flags a page, that page becomes ineligible for ranking in Discover. Content that clears both filters stays in the eligible pool. Crucially, this filtering step happens before the computation-heavy ranking stage, which the paper says also reduces the computing resources Google spends on ranking content that was never going to surface anyway.
The numbers behind the disclosure
The research paper is light on raw data points, but the ones it does share are notable for a two-year-old live system getting its first public explanation:
- Two years of deployment before Google said anything about it publicly, via a research paper rather than a product announcement or help-center update.
- 54.9% reduction in user complaints about stale content in Discover over that two-year span.
- Two distinct filtering stages, run before the main ranking computation.
- Alongside the complaint reduction, Google reported “modest gains” in engagement and improved content diversity in Discover feeds, though the paper does not attach a specific percentage to either of those two claims.
Search Engine Journal’s reporting is careful to note that the paper names no individual researchers in quotes beyond describing the system itself, and that Google has not issued a separate statement confirming the findings through its usual search liaison channels. The primary source, in other words, is the paper itself. It’s also the latest example of Google explaining a live system well after the fact, a pattern we tracked recently in Google’s October AI Mode and AI Overviews link changes.
Why this matters for publishers right now
Freshness has been treated as a soft ranking signal for years, something SEOs nod along to without much specificity. SDF turns it into something closer to a gate: a page that trips either filter doesn’t get a chance to compete in ranking at all, in Discover specifically.
That has a few concrete implications. A content library’s mix matters more than any single post’s freshness, which is exactly the kind of audit our content marketing team runs before touching a single headline. Evergreen content that genuinely holds up over time should pass Relevance Decay cleanly. Time-limited content, pricing, seasonal guides, anything tied to a specific offer or tool version, needs an explicit update or retirement plan, or it risks tripping the decay filter on a predictable schedule. And content that covers a fast-moving topic needs to be checked against what’s since superseded it, not just against its own publish date.
Montti’s reporting also raises a practical wrinkle: publication date alone is an unreliable proxy for either kind of staleness. A two-year-old evergreen piece can be perfectly fresh by SDF’s standard, while a post published last month can already be decaying if the space around it has moved fast.
ARC Marketing’s Take
“Google telling us a stale-content filter has been live for two years isn’t really news about an algorithm, it’s news about how much of ranking we’ve already lost visibility into,” notes ARC Marketing’s SEO team. “The real takeaway for site owners is that ‘go update your old content’ stopped being generic advice and became a measurable ranking signal with its own name.”
ARC Marketing has started running quarterly “supersession audits” on every client’s top 50 evergreen pages: flagging any page where a competitor has visibly published a more current version of the same topic, so it gets refreshed before it’s filtered out of Discover rather than after traffic has already quietly dropped off. It’s the same granular, algorithm-level diligence that underpins ARC Marketing’s GEO & AEO work for clients.
Sources
- Roger Montti, “Details Emerge About Google Discover’s SDF Freshness Algorithm,” Search Engine Journal, October 21, 2026.
- Google Research, “Decomposing Staleness in Recommender Systems: A Dual-Filter Framework for Supersession and Decay,” arXiv.
Frequently Asked Questions
What is Google’s SDF freshness algorithm?
SDF stands for Supersession-Decay Filtering, a system Google uses in Google Discover to filter out content it judges to be stale before that content ever reaches Discover’s main ranking stage. It was detailed in a Google research paper reported on by Search Engine Journal on October 21, 2026.
How long has SDF been running in Google Discover?
According to the research paper, SDF has been deployed in Discover for two full years. Google had not previously disclosed the system publicly; it surfaced through the paper rather than an official announcement.
What’s the difference between supersession and relevance decay?
Supersession is when newer, more current content makes an older page outdated. Relevance Decay is when a page becomes less useful over time on its own, even if nothing newer has replaced it. SDF uses one filter for each.
Does SDF affect regular Google Search results, or just Discover?
The research paper and Search Engine Journal’s reporting describe SDF specifically as a system deployed in Google Discover. Neither source states that SDF applies to core Google Search ranking.
What results has Google reported from SDF?
Over its two-year deployment, Google’s paper reports a 54.9% reduction in user complaints about stale content in Discover, along with modest engagement gains and improved content diversity. The paper does not provide numeric detail for the engagement or diversity improvements.
Where did these details about SDF come from?
They come from a Google research paper titled “Decomposing Staleness in Recommender Systems: A Dual-Filter Framework for Supersession and Decay,” published on arXiv and reported on by Roger Montti at Search Engine Journal on October 21, 2026.