How NavBoost Actually Decides Who Ranks First

Google spent nearly two decades denying that clicks influenced rankings. Engineers dismissed the idea publicly. Spokespeople called it "made up crap." Then in October 2023, during the DOJ antitrust trial, Google VP Pandu Nayak took the stand and confirmed what many SEOs had suspected: NavBoost, a system built in 2005, is "one of the most important" ranking signals Google uses. It stores 13 months of click data on every search result you interact with, and it uses that data to decide which pages deserve to rank.

This was not a small admission. NavBoost is referenced 84 times in the leaked Google API documentation from 2024, with five separate modules built around it. The system does not care about your keyword density or your backlink profile. It cares about what real users do after they click.

What NavBoost Is and How It Processes Search Queries

NavBoost is a click-based ranking system that collects and stores user interaction data from Google search results over a rolling 13-month window. Prior to 2017, that window was 18 months. Every search you perform, every result you click, every time you bounce back to try a different link, NavBoost records it.

The system does not function as an initial retrieval mechanism. As Nayak stated during his testimony, "you get NavBoost only after they're retrieved in the first place." Google's other systems pull thousands of candidate documents for a query. NavBoost then narrows that set from tens of thousands to a few hundred based on historical user behavior. Those refined candidates get passed to Google's machine learning ranking systems for final ordering.

This is where most explanations of NavBoost stop. But the system goes deeper.

The NavBoost Click Taxonomy and How SEO Changes

Not all clicks are equal in NavBoost's evaluation. The 2024 API documentation leak revealed that Google classifies clicks into distinct categories, each stored as a separate float-valued attribute:

goodClicks represent interactions where the user clicked a search result and stayed. The exact threshold is not public, but the pattern is clear: the user found what they needed and did not return to the SERP.

badClicks are the opposite. The user clicked, arrived at the page, and bounced back to Google quickly. This is the behavior Google spent years claiming had no impact on rankings. The leaked documentation proves it is a named, stored, and weighted signal.

lastLongestClicks track the final result a user visited in a search session where they also spent the most time. If someone searches "navboost," clicks three results, and spends eight minutes reading the third page before closing their browser, that third page earns a lastLongestClick. This is one of the strongest positive signals in the system. Cyrus Shepard at Moz has written extensively about how first, long, and last clicks function as engagement signals in Google's ranking systems.

unicornClicks remain the most mysterious metric. The leaked documentation does not define them explicitly. They likely represent exceptionally rare interactions where user satisfaction was obvious and unambiguous, something well beyond a standard goodClick.

What makes this system sophisticated is the squashing function. NavBoost does not use raw click counts. It applies a normalization process, described in a Google patent on local search scoring, that compresses large values logarithmically. Without squashing, a BBC article with 50,000 clicks would permanently outrank a specialist blog with 500 clicks, even if the specialist blog had a far better ratio of goodClicks to badClicks. Squashing lets the ratios speak. The leaked API includes both "squashed" and "unsquashed" versions of each click metric, and a large disparity between the two can flag a result for additional scrutiny, likely as an anti-manipulation measure.

How NavBoost, Glue, and Slices Segment Search Data

NavBoost handles the traditional "ten blue links" in search results. A companion system called Glue handles everything else on the SERP: knowledge panels, image carousels, video results, and People Also Ask boxes. Glue tracks clicks, hovers, scrolls, and swipes on these elements and creates a common metric to compare them with standard web results.

During the antitrust trial, Nayak explained that NavBoost operates early in the ranking pipeline, narrowing candidate documents. Glue operates later, determining which SERP features to display and where to place them. Both systems feed into Google's broader assessment of user satisfaction.

NavBoost also slices its data along two dimensions: device type and geographic location. Your mobile rankings and desktop rankings are influenced by separate engagement pools. A page that performs well on mobile in Texas may rank differently on desktop in New York, because NavBoost maintains distinct behavioral profiles for each combination. This is why we see different ranking behavior across devices for the same queries, even when the content is identical.

The slicing mechanism also explains why local search results can shift rapidly. When NavBoost collects enough location-specific interaction data for a query, it can override broader national patterns. A regional service provider with strong local user engagement can outperform a national brand in search results for that geographic slice, even if the national brand has more total clicks.

The Google Patent Behind NavBoost

The conceptual blueprint for NavBoost appears in US Patent 8595225B1, titled "Systems and methods for correlating document topicality and popularity." Filed in September 2004 by Amit Singhal and Urs Hoelzle, the patent describes a method for assigning "popularity scores" to documents based on user navigation patterns and then correlating those scores with specific topics.

The patent does not mention clicks by name. Instead, it describes "user navigational patterns" and "documents visited by users," which are functional equivalents. The filing date aligns with Nayak's testimony that NavBoost launched in 2005. Search Engine Journal was among the first to draw the connection between this patent and the NavBoost system revealed in the trial.

What the patent describes, and what the leaked API documentation confirms, is a system that moved Google beyond simple link-based authority signals. NavBoost introduced a feedback loop where the users themselves vote on which results deserve to rank, and those votes carry weight for 13 months.

What NavBoost Means for SEO Strategy and Search Results

If you work in technical SEO, NavBoost changes the optimization equation. Ranking is no longer just about matching keywords and building links. It is about earning clicks and then keeping users on your page.

We have applied this understanding across client campaigns in Nashville and beyond. The implications break down into measurable actions:

Earn the click. Your title tag and meta description are NavBoost's entry point. If users do not click your result, NavBoost has no positive data to store. We have seen pages with strong content stall in rankings because their SERP snippets failed to compel clicks. Title optimization is not vanity work. It is the first gate in the NavBoost pipeline.

Reduce badClicks. Every pogo-stick, every quick bounce back to the SERP, registers as a negative signal that persists for 13 months. Page speed, content relevance, and answering the user's actual question within the first few paragraphs all reduce badClicks. We wrote about how Google's click signals feed into ranking decisions, and NavBoost is the system that stores those signals.

Earn lastLongestClicks. The page where users spend the most time and then stop searching is the one NavBoost rewards most heavily. Depth matters. Structure matters. Giving users a reason to scroll, read, and stay is the single best thing you can do for NavBoost performance. This connects directly to what Google's Helpful Content classifier evaluates: whether your page actually satisfies the search.

Build brand recognition. Tom Capper at Moz showed that branded search volume is a stronger predictor of organic rankings than Domain Authority. NavBoost helps explain why. When users search for your brand and click your result consistently, NavBoost stores that as a strong positive signal. That signal then bleeds into non-branded queries where you compete. Brand is not separate from SEO. Brand is NavBoost data.

The rendering and crawling work we covered in our post on closing the rendering gap matters here too. If Google cannot render your page properly, users who click through will encounter broken layouts, triggering badClicks that damage your NavBoost profile.

For a broader view of how these signals connect to Nashville businesses specifically, our Nashville SEO Playbook covers the local slicing dimension of NavBoost and why geographic user data matters for local rankings.

The NavBoost System Google Built in Secret

Google built NavBoost in 2005 and kept it hidden for 18 years. When it finally surfaced through court testimony and leaked documents, it confirmed what the SEO industry had long suspected: user behavior directly influences rankings. The clicks you earn, the time users spend, and the satisfaction they express through their behavior are all stored, weighted, and applied to every future search.

NavBoost is not a peripheral signal. It is foundational. And the 13-month memory means that every improvement you make today compounds for over a year.

The question is no longer whether user interactions with search results matter for rankings. The question is whether your pages earn the right clicks, keep users engaged, and satisfy the queries they arrived with.