Google's Glue system re-ranks results using SERP behavior you can't see

What happened

SALT.agency has published an analysis of Google’s internal “Glue” system, piecing together references from the US Department of Justice antitrust case and leaked internal documents. Their reading is that Glue is a “Super Query Log” that captures user behavior across SERP features beyond the traditional blue links.

In SALT.agency’s account, where NavBoost tracks clicks on standard organic results, Glue monitors micro-interactions with Universal Search features: Map Packs, Video Carousels, Featured Snippets, and Image Boxes. The interactions it captures include hovers, scrolls, swipes, and dwell time on specific modules, even when no click occurs.

According to the analysis, Glue and NavBoost feed into what Google calls the Popularity Signal. Every query has a Popularity Score built from 13 months of behavioral data. The score is per query, not per page, so a page’s behavioral performance is weighed against competing results for that specific query. Positive signals include “good clicks” (the user stays on the clicked site) and “longest clicks” (the last site visited before ending the search session).

Negative signals include what Google internally calls “bad clicks,” meaning short dwell-time visits where a user clicks through and quickly returns to the SERP. The concept overlaps with what the SEO industry has long called pogo-sticking, though Google’s internal definition may be narrower. SALT.agency’s analysis also suggests that impressions without clicks count against a result. SALT.agency notes that the weightings of these interactions are not known, as Google’s systems have a high level of imponderable complexity.

In SALT.agency’s interpretation, Google first applies traditional quality signals like keywords and links to produce an initial results list, then applies the Popularity Signal to re-order it. Under this model, a page in position five with a higher Popularity Score than the page in position one could get swapped above it.

Why it matters

The SEO industry has debated whether pogo-sticking and SERP behavior affect rankings for over a decade. Google has historically downplayed the role of click data in ranking. If SALT.agency’s reading of the leaked documents is accurate, Glue would represent a more systematic behavioral re-ranking layer than most practitioners assumed existed. And most of it would be invisible: Search Console reports impressions and clicks, but not hovers, scroll-past rates, or dwell time on SERP features.

Under SALT.agency’s model, results that accumulate lower-than-expected CTR for their position and query send negative behavioral signals. Those signals contribute to demotion over time. Demotion reduces future impressions, which reduces chances to recover. Under this framing, the decline would be purely algorithmic, with no manual penalty to appeal.

SALT.agency also claims that mobile users consistently scroll past text results to reach the Map Pack on local queries, and suggests Glue may capture this scroll behavior as a reordering signal. Map Pack positioning is influenced by multiple factors including query intent and device type, so how much behavioral data adds on top is unclear.

The analysis claims a 13-month rolling window for behavioral data, which would mean CTR improvements work against a long baseline. The source does not describe how recent signals are weighted against older ones within that window.

SALT.agency frames brand recognition as a self-reinforcing advantage: users who recognize a brand click more, which feeds Glue and NavBoost, which pushes the brand higher. They call this a “defensive moat.” Brand advantages in search are real and predate any Glue-specific mechanism, but if behavioral data feeds back into rankings at the scale SALT.agency describes, the incumbency advantage gets harder to overcome.

What to do

None of the advice below depends on Glue being real. Pogo-sticking and CTR have been practical ranking concerns for years, and these are good hygiene regardless.

Audit your impression-to-click ratio in Search Console for high-impression queries. Pages with thousands of impressions and lower-than-expected CTR are worth investigating whether or not a formal Popularity Score exists. Prioritize these for title tag and meta description rewrites that better match search intent.

Reduce back-button bounces on your top-ranking pages. If users click through and immediately return to the SERP, that has always been a bad signal. Check whether your landing pages deliver on the promise of the title tag and snippet. Slow load times, aggressive interstitials, and content that does not match search intent all drive pogo-sticking. (Google is also enforcing against deliberate back-button hijacking as a separate initiative.)

For sites competing in SERP feature spaces (video carousels, featured snippets, image boxes), track whether your features are holding position over time. Use Search Console’s search appearance filters and third-party rank trackers that log SERP feature ownership to spot losses. If your featured snippet disappears or your video carousel slot drops, investigate whether the content still matches the query intent before assuming an algorithmic cause.

SALT.agency’s core argument is that “SEO can’t fix a broken brand and user experience.” That has always been true. If Glue works as described, it would formalize what practitioners have long suspected: user satisfaction signals feed back into rankings, and no amount of link building overcomes a page users do not want to stay on.