What Glue Is
Modern Google search results are rarely a simple list of ten blue links. A typical query returns a blended page that may include a featured snippet at the top, a People Also Ask block, an image or video carousel, a knowledge panel in the sidebar, a local pack with a map, and several organic results interleaved among them. Something has to decide which of these elements appear, in what order, and how prominently. According to testimony from Google's own Vice President of Search, that something is called Glue.
Glue is the system Google uses to assemble the universal — or blended — search results page using real user-interaction data. During the 2023 United States antitrust trial against Google, Pandu Nayak described Glue as "another name for NavBoost that includes all of the other features on the page." In other words, Glue is closely related to the click-based re-ranking system this site documents in detail, but its scope is the whole page rather than just the organic web listings.
The distinction matters because it reframes what user behavior influences. It is not only the ordering of ten organic results that responds to clicks. The presence and placement of nearly every visible element on a results page appears to be informed by how people have interacted with similar layouts in the past. For a foundational overview of the underlying click system, see What is NavBoost?
What follows draws on three primary sources: sworn testimony from the antitrust trial, the May 2024 leak of internal Google API documentation, and the technical analyses that followed both. Where Glue's internal mechanics are not directly confirmed, the discussion below hedges accordingly. Glue is one of the less thoroughly documented systems in Google's stack, and several details remain inferred rather than established.
NavBoost vs. Glue: The Division of Labor
The cleanest way to understand Glue is by contrast with NavBoost. The two systems share a signal family and a common purpose — using aggregated user behavior to improve the results page — but they operate on different parts of that page.
NavBoost is concerned with the traditional organic web results, the so-called ten blue links. It re-ranks those listings based on how users click and behave after clicking, using the click types revealed in the 2024 API leak. For a detailed breakdown of those categories, see NavBoost Click Types, and for the end-to-end pipeline, see How NavBoost Works.
Glue extends that logic to everything else. As one analysis of the leak summarized it, where NavBoost focuses only on organic search results, Glue includes NavBoost plus everything in the results page that is not an organic web result — Google Ads, map packs, knowledge panels, images, and the rest. Glue is the broader system responsible for managing and ranking the rich results and interactive elements that surround the organic listings.
| Dimension | NavBoost | Glue |
|---|---|---|
| Scope | Organic web results (ten blue links) | The whole blended page, including non-web features |
| What it orders | Ranking of organic URLs | Which features appear and where they sit |
| Primary signals | Clicks and post-click behavior | Clicks, hovers, scrolls, swipes |
| Output | Re-ranked organic list | Assembled universal SERP |
| Real-time variant | Aggregated over a long window | Glue plus InstantGlue (reported ~24h) |
The relationship is best understood as nested rather than parallel. Nayak's framing — Glue as "another name for NavBoost that includes all of the other features" — suggests Glue is not a wholly separate system bolted on beside NavBoost, but an expansion of the same approach to a larger surface. The organic re-ranking that NavBoost performs is, in this view, one component of the larger page-assembly job that Glue oversees.
Key distinction
NavBoost answers "which organic results should rank highest?" Glue answers "what should the entire page look like?" The same user-behavior signals inform both, but Glue's output is a full-page layout, not just an ordered list of links.
The Signals Glue Uses
The most specific public description of Glue's inputs comes from the antitrust trial. According to that testimony, Glue aggregates diverse types of user interactions — clicks, hovers, scrolls, and swipes — and converts them into a common metric that allows web results and search features to be compared on a single scale. That common-metric concept is central to how Glue can work at all.
Why a Common Metric Is Necessary
Comparing a web result to a video carousel to a knowledge panel is not straightforward, because users interact with each differently. Nobody "clicks through" a knowledge panel the way they click a blue link; they read it in place, or hover over a sub-element, or scroll past it. A featured snippet may satisfy a query entirely without any click at all. To decide whether a snippet or a carousel deserves a prominent slot, Google needs a way to put these dissimilar interactions on comparable footing.
That is the role of the common metric. By translating clicks, hovers, scrolls, and swipes into a unified engagement score, Glue can ask a single question of every candidate element: how much does this satisfy users for this query, relative to the alternatives competing for the same space? The element that earns the most genuine engagement, in this framing, earns its position.
Beyond the Click
The inclusion of hovers, scrolls, and swipes is notable because it extends user-behavior signals past the click event that dominates NavBoost discussions. On a results page dense with features, much of the meaningful interaction is non-click: a user hovering over an image in a pack, swiping through a carousel on mobile, or scrolling past a People Also Ask block without expanding it. These behaviors carry information about what users find useful, and Glue appears designed to capture it.
This is also where the engagement-versus-analytics distinction matters. The behaviors Glue reads are interactions with Google's own results page, captured at the SERP level — not metrics drawn from third-party analytics tools. Google has stated repeatedly that it does not use Google Analytics or GA4 engagement data as a ranking input. The signals Glue uses are first-party SERP interactions, which is a different category entirely. For more on how these signals relate to ranking, see How SERP Features Change CTR.
What the Leak and Trial Revealed
Public knowledge of Glue rests primarily on two events: the 2023 antitrust trial and the 2024 API documentation leak. Neither produced a complete specification, but together they established that Glue exists, what it broadly does, and how it relates to the rest of Google's ranking infrastructure.
The Antitrust Trial
In United States v. Google LLC, the trial that ran from September to November 2023, Pandu Nayak testified about Google's ranking systems under oath. His description of NavBoost as one of the most important ranking signals drew the most attention, but his characterization of Glue as the page-wide extension of NavBoost was equally significant. It was the clearest official confirmation that user-interaction data shapes not just organic rankings but the assembly of the entire results page. The detail on this point in this article is drawn from that trial testimony.
The 2024 API Leak
The leak of roughly 2,596 internal API modules and more than 14,000 attributes in May 2024 — disclosed publicly by Rand Fishkin of SparkToro and analyzed technically by Mike King of iPullRank — added structural context. The documentation referenced the click fields that power NavBoost and surfaced the architecture in which Glue operates. For the full account of what surfaced, see The 2024 Google API Leak.
The RESONEO Analysis and System Architecture
Among the most detailed independent breakdowns is the RESONEO series, whose fifth installment ("Click-data, NavBoost, Glue, and Beyond") placed Glue within Google's broader serving architecture. According to that analysis, NavBoost and Glue operate within a component called Tangram at the Superroot level — the layer where Google's re-ranking mechanisms sit — and apply their adjustments after the main initial ranking system (Ascorer) has produced a first ordering. Tangram is described as a puzzle-assembly system that integrates images, videos, maps, news, and other features into the final page with the help of Glue and NavBoost.
The Tangram, Superroot, and Ascorer terminology comes from third-party interpretation of leaked documentation rather than confirmed Google statements. These names describe the architecture as reconstructed by analysts; the precise responsibilities of each component should be treated as informed inference, not established fact.
How Glue Decides Which Features Appear
The practical question for anyone working in search is how Glue's signals translate into the layout a user actually sees. The available evidence supports a general model, even if the exact thresholds and weightings are not public.
Triggering and Positioning
Glue's common metric appears to govern two related decisions: whether a feature triggers for a given query at all, and where on the page it is placed if it does. A featured snippet, for example, is not shown for every query. Its appearance depends in part on whether the engagement data indicates that users find a snippet helpful for that query type. The same logic plausibly applies to image packs, video carousels, and People Also Ask blocks.
Once a feature is eligible to appear, its position is a competitive matter. Every element vies for the limited prime real estate near the top of the page. The common metric lets Glue rank these heterogeneous candidates against one another, so that the feature earning the strongest engagement for the query rises, and weaker candidates fall or are dropped. For featured snippets specifically, this competition has direct CTR implications, covered in Featured Snippets and CTR.
The Feedback Loop
This creates a self-reinforcing cycle. When users consistently engage with a particular feature for a particular query — clicking the snippet, swiping the carousel, expanding a PAA answer — Glue's metric for that feature strengthens, making it more likely to appear and to appear prominently. When a feature is consistently ignored, its metric weakens. Over time, the blended page for any given query tends toward the layout that historical users have found most satisfying.
This is the same evidence-first principle that underlies NavBoost: the system does not assume which features are best in the abstract; it observes which ones users actually prefer and adjusts accordingly.
Relationship to Twiddlers
Glue does not act alone in shaping the final page. The leak analyses describe a layer of re-ranking functions known as twiddlers that adjust results after initial scoring, often to enforce specific rules or boost particular categories of content. Glue and NavBoost are part of this re-ranking ecosystem, supplying the user-engagement signals that some twiddlers act on. For how these adjustment functions fit together, see Twiddlers: How Google Re-Ranks Results.
InstantGlue and Real-Time Pages
Standard Glue, like NavBoost, aggregates interaction data over a long window, which makes it stable but slow to react. That is a problem for breaking news and rapidly emerging queries, where the right page layout can change within hours. The leak analyses describe a real-time counterpart built to address this: InstantGlue.
According to that analysis, InstantGlue works alongside real-time boost mechanisms and targets emerging queries and breaking news content. It reportedly collects the same user-interaction signals as Glue but processes only roughly the last 24 hours of log data, refreshing approximately every 10 minutes. This lets the page assembly respond quickly: when a story breaks and users suddenly start engaging heavily with news results or a fresh video carousel, InstantGlue can surface those features far faster than the long-window system would.
The roughly 24-hour window and 10-minute refresh cadence attributed to InstantGlue come from third-party leak analysis, not confirmed Google documentation. They are reported here as the best available estimate, and the exact figures should be treated as approximate.
The pairing of a slow, stable system with a fast, reactive one is a recurring pattern in Google's infrastructure. The long-window version provides durable, manipulation-resistant rankings, while the real-time version handles the volatility of fresh events. Together they let the blended page be both stable for established queries and responsive for breaking ones.
SEO Implications
If Glue governs which features appear and where, then visibility on a modern results page depends on more than ranking a URL well. Winning a featured snippet slot, a spot in a People Also Ask block, or placement in an image or video carousel is, in this model, partly a function of Glue-positive engagement.
Earning and Keeping Feature Slots
The implication is that a feature slot is not held permanently by whoever first earns it. If users consistently engage with a snippet sourced from a particular page, that engagement reinforces the slot. If a different source begins earning stronger engagement, the slot can shift. The same competitive dynamic that NavBoost applies to organic rankings appears to apply, via Glue, to feature placement.
This reframes feature optimization. Earning a snippet is not only about formatting content so Google can extract it; it is also about producing an answer that users find genuinely satisfying once it appears, so that the engagement signals sustain the placement. Content that wins a slot but disappoints users may not keep it.
Engagement Quality Over Engagement Volume
Because Glue shares NavBoost's normalization and aggregation defenses, the engagement that matters is genuine and sustained, not artificially inflated. A burst of automated interaction faces the same diminishing returns and the same long-window dilution that limit click manipulation against organic rankings. This is one reason the practitioners who study engagement-based ranking emphasize real human behavior over synthetic signals — and why services such as SerpClix, which is built around a crowd-sourced pool of more than 400,000 real human clickers rather than automated traffic, are sometimes referenced in discussions of genuine engagement. The relevant point for Glue is the same as for NavBoost: durable outcomes come from features users actually prefer.
Designing for Multiple Interaction Types
Glue's use of hovers, scrolls, and swipes alongside clicks suggests that engagement should be considered across interaction types, not just clicks. On mobile especially, swipe-through behavior on carousels and scroll patterns past feature blocks may carry weight. The general guidance that follows from the evidence is to ensure that whatever feature a page competes for actually rewards the interaction it invites — a carousel image that earns the swipe, a snippet that earns the read.
Shared Manipulation Resistance
Because Glue is built on the same signal family as NavBoost, it inherits the same structural defenses against manipulation. Understanding these defenses clarifies why short-term tactics tend not to produce durable changes in page layout.
| Defense | Effect on manipulation |
|---|---|
| Signal normalization | Compresses extreme interaction volumes so a burst of engagement yields diminishing returns rather than a proportional boost |
| Long aggregation window | Dilutes short-term activity against many months of genuine historical interaction |
| Common-metric comparison | Forces every feature to compete on real engagement against alternatives, limiting any single inflated element |
| Multi-signal correlation | Cross-references engagement against other quality indicators to flag inconsistent patterns |
The normalization mechanism in particular mirrors the squashing function that protects NavBoost's organic rankings: extreme signal volumes are compressed so that no single flood of activity dominates. The practical consequence is that influencing which features appear, like influencing organic rank, rewards sustained genuine engagement over short bursts of artificial activity.
The most reliable way to influence a feature's placement, on the available evidence, is to make the feature genuinely useful for the query it serves. Engagement that reflects real user satisfaction is precisely the signal Glue is designed to reward, and it is also the signal hardest to fake at scale.
Open Questions and Limits of the Evidence
Glue is less thoroughly documented than NavBoost, and several aspects remain uncertain. The common metric is described in testimony but never quantified; the exact way clicks, hovers, scrolls, and swipes are weighted relative to one another is not public. The InstantGlue figures are reported through leak analysis rather than confirmed. The Tangram and Superroot architecture is a reconstruction by analysts, not an official map.
What is well supported is the core claim: Google uses aggregated user-interaction data not only to rank organic results but to assemble the entire blended page, and the system that does the latter is Glue. Google's own Search Liaison, Danny Sullivan, acknowledged after the leak that Google uses "a variety of different ranking signals including, but not solely, aggregated and anonymized interaction data." That statement is consistent with a page-assembly system driven in part by user behavior, even as it stops short of confirming Glue's internals.
For readers tracing the broader system, the companion references on this site cover the organic side in depth: start with What is NavBoost? for the foundation, then move to the technical architecture and click-type detail.
Frequently Asked Questions
What is Glue in Google Search?
Glue is the user-interaction system Google uses to assemble and order the full search results page, including features that are not standard organic web results. During the 2023 antitrust trial, Pandu Nayak described it as another name for NavBoost that includes all of the other features on the page. Where NavBoost re-ranks the ten blue links, Glue aggregates clicks, hovers, scrolls, and swipes across the whole page to decide which SERP features appear and where they sit.
How is Glue different from NavBoost?
NavBoost re-ranks standard organic web results based on click behavior. Glue applies the same family of interaction signals to everything else on the page, such as featured snippets, People Also Ask, knowledge panels, image and video carousels, and local packs. NavBoost handles the ten blue links; Glue handles the blended or universal SERP. The two are closely related and, by Nayak's own description, Glue is essentially NavBoost extended to all page features.
What signals does Glue use?
According to antitrust trial testimony, Glue aggregates diverse types of user interactions including clicks, hovers, scrolls, and swipes, and converts them into a common metric so that web results and search features can be compared on the same scale. This lets Google decide whether a feature such as a featured snippet or video carousel should trigger for a query, and where on the page it should appear.
What is InstantGlue?
InstantGlue is described in third-party analysis of the 2024 API leak as a real-time counterpart to Glue aimed at emerging queries and breaking news. The evidence suggests it collects the same interaction signals as Glue but processes only roughly the last 24 hours of log data and updates approximately every 10 minutes, allowing the page layout to react quickly to fast-moving events. These figures come from leak analysis rather than confirmed Google documentation, so they should be treated as reported rather than certain.
Does Glue affect featured snippets and People Also Ask?
The available evidence indicates that Glue helps determine which SERP features are shown and how they are positioned, which would include featured snippets and People Also Ask boxes. Because Glue uses engagement signals, a snippet or PAA answer that consistently earns clicks and dwell is more likely to keep its slot, while one that is repeatedly skipped may be demoted or removed. The precise internal logic is not confirmed, so this should be read as a reasonable inference from testimony and leak analysis.
Can Glue signals be manipulated?
Glue inherits the same manipulation defenses as NavBoost, including normalization of extreme signals and long aggregation windows, so artificially inflating engagement with a single feature faces the same diminishing returns. Genuine, sustained user interaction is far harder to fake than a short burst of automated activity, which is why services that rely on real human users, such as SerpClix, are sometimes discussed in this context. The evidence still indicates that durable layout outcomes come from features users actually prefer.
Further Reading
- What is NavBoost? — the cornerstone overview of Google's click-based re-ranking system that Glue extends to the whole page.
- How NavBoost Works — the end-to-end technical architecture, from click collection through normalization and aggregation.
- NavBoost Click Types — the goodClicks, badClicks, and lastLongestClicks categories revealed in the API leak.
- Twiddlers: How Google Re-Ranks Results — the re-ranking functions that operate alongside NavBoost and Glue.
- How SERP Features Change CTR — how the page features Glue assembles affect click-through rates by position.
- The 2024 Google API Leak — the documentation that surfaced the architecture in which Glue operates.