NavBoost: How Google Uses Click Behavior to Re-Rank Search Results

01What is NavBoost?

NavBoost is Google's click-based re-ranking system. It observes what users do after a Google search — which results they click, how long they stay, and whether they return to the search page. Google uses this behavior to adjust rankings. NavBoost is not a content analysis system and does not read web pages. It treats user behavior as a proxy for content quality: if users consistently click a result and stay, that result probably satisfied their need.

Google's Vice President of Search, Pandu Nayak, testified under oath during the US Department of Justice antitrust trial in 2023 that click data plays a central role in search ranking. This confirmed NavBoost as one of Google's "most important" ranking signals. It was the first official confirmation from a senior Google executive that click behavior is used as a direct ranking signal. It contradicted years of public statements from Google that downplayed the role of click-through rate.

02The click signals: goodClicks, badClicks, lastLongestClicks

The Google API leak of May 2024 revealed the fields NavBoost uses to classify clicks. Three field names appear in the leaked documentation: goodClicks, badClicks, and lastLongestClicks. They are not raw click counts. They are classified outcomes of user interactions.

goodClicks are clicks where the user appeared satisfied: they clicked a result and did not return to the search page, or they spent a meaningful amount of time on the destination. badClicks are the opposite: the user clicked through but quickly bounced back to the search results, which indicates the page did not meet their need. lastLongestClicks is the strongest signal of all. It captures the last, longest click in a search session and marks the result that finally satisfied the user's need.

The line between good and bad clicks is not binary. A squashing function, which compresses raw values into a bounded range, weights clicks by dwell time, click position, and other factors. Google engineers set the exact thresholds by hand. They are not machine-learned. That means they can be analyzed, adjusted, and debugged. The DOJ trial revealed this to be a deliberate architectural choice at Google.

03The 13-month aggregation window

NavBoost does not work in real time. It aggregates click behavior over a rolling 13-month window. A page's NavBoost score therefore reflects a full year of user interaction data, weighted toward more recent behavior. The long window makes NavBoost resistant to short-term manipulation: a few days of artificial clicks will not meaningfully move a score built on months of genuine user behavior.

The DOJ trial also revealed a companion system called "Instant Glue" that operates on a 24-hour log of user interaction data. Instant Glue provides near-real-time freshness boosts for queries where timeliness matters. It is a separate mechanism from the 13-month NavBoost aggregation. The two systems complement each other: NavBoost provides the stable, long-term quality signal, and Instant Glue provides the short-term freshness signal.

04The DOJ trial: what was revealed under oath

The US v. Google antitrust trial (September–November 2023) produced the most significant disclosures about Google's ranking systems in the company's history. For years, Google maintained a carefully cultivated ambiguity about the role of user behavior signals in ranking. When SEO practitioners observed that click-through rate appeared to affect rankings, Google could answer with technically accurate but misleading statements about how CTR was "not a ranking factor." The trial made that ambiguity untenable.

Pandu Nayak, Google's VP of Search, confirmed under oath that click data is used as a direct ranking signal. He described a system that tracks which results users click on, how long they stay, and whether they return to the search results page. This contradicted earlier public statements from Google employees. Matt Cutts, the former head of web spam, had repeatedly stated that click-through rate was too noisy to serve as a reliable ranking signal.

A later exhibit, PXR0356, documented a February 2025 call with Google engineer Hyung-Jin Kim. Kim confirmed that NavBoost was his second signal project at Google, that he has many patents related to it, and that he spent many years developing it. The exhibit links the leaked system name "NavBoost" directly to a specific Google engineer with patent filings.

05Patent analysis: three Stufe 1 patents

Three US patents directly name Hyung-Jin Kim as inventor and describe systems for modifying search result ranking based on implicit user feedback, which is the mechanism NavBoost implements. All three are classified as Stufe 1 (the strongest evidence bridge) because DOJ Exhibit PXR0356 confirms the inventor as the NavBoost developer.

US10229166B1, granted March 2019, describes "modifying search result ranking based on implicit user feedback" — the core NavBoost mechanism. US8661029B1, granted February 2014, is the original patent in the family and was filed in 2006. US8938463B1, granted January 2015, extends the system to correct for presentation bias, which is the tendency of users to click higher-ranked results regardless of quality.

A fourth patent, US9092510B1, was reviewed and discarded. It describes a temporal element of user feedback, but its inventors (Stets, Paskin) have no connection to Kim. That makes it Stufe 3, functional similarity only. The case shows the source discipline that distinguishes this reference from typical SEO commentary: a patent that sounds relevant is not included unless the evidence chain is strong enough.

06NavBoost in the ranking architecture

Google's ranking ultimately boils down to two fundamental top-level signals, as the DOJ trial revealed: Quality (Q*) and Popularity (P*). NavBoost feeds into the Popularity signal, which measures how widely visited and well-linked a page is. Quality (Q*) is the other pillar. It assesses the trustworthiness and authority of a website and is heavily influenced by PageRank.

NavBoost is part of the ABC signals — Anchors (links), Body (content terms), and Clicks (user behavior). Together they form the base topicality score (T*) and determine how relevant a document is to a query. NavBoost's specific contribution is the click component: it provides the behavioral verification that a document actually satisfies users for a given query.

NavBoost feeds into GLUE (System 3), which combines NavBoost with Chrome visit data into a unified user behavior signal. The relationship is hierarchical: NavBoost captures search-driven clicks, GLUE adds direct-visit data from the Chrome browser, and together they form a comprehensive picture of user satisfaction.

07Implications for SEO practitioners

NavBoost cannot be directly manipulated. It reflects genuine user satisfaction aggregated over 13 months. This is by design: Google wants a signal that resists gaming. The implications for SEO are far-reaching.

First, title and meta description optimization matters more. These elements determine whether users click and whether the page they reach meets the expectation the snippet set. A page that ranks well but generates badClicks, because its snippet overpromises or its content underdelivers, will see its ranking degrade over time as NavBoost accumulates negative behavioral data.

Second, 'pogo-sticking' (clicking a result, bouncing back, clicking another result) is confirmed as a real negative signal. NavBoost's badClicks classification captures this pattern directly. Pages that consistently fail to satisfy users will be demoted, even if they rank well at first through other signals.

Third, the long aggregation window means SEO changes take time to show their full effect. A content improvement made today will not fully influence NavBoost for months. This is why Google's ranking changes often appear gradual rather than sudden: the behavioral signal needs time to accumulate.

Finally, the DOJ trial confirmed that almost all Google signals are hand-crafted (not ML-based). The ranking system is therefore ultimately understandable by human engineers. This is a significant finding for the SEO industry: systematic observation, hypothesis testing, and evidence-based analysis can uncover how the system works, which is exactly what this reference aims to do.

08References