NavBoost: What Google's Click System Means

NavBoost is a Google ranking system that uses how people click on search results. Documents from the US v. Google antitrust case describe it as “one of the most powerful ranking components” at Google, remembering clicks for every query over the previous 13 months, broken down by location and device. The 2024 documentation leak added the names of the click attributes Google stores, including goodClicks, badClicks and lastLongestClicks.

This page is for SEOs and marketers who want the primary sources rather than the retellings. It quotes the court documents and the leaked module files directly, and then turns them into practical advice on titles, snippets and satisfaction.

On this page

• What is NavBoost?
• What the DOJ case revealed about NavBoost
• The click attributes in the leaked documents
• What is a last longest click?
• Why fake clicks are a bad bet
• What NavBoost means for titles and snippets
• Working with NavBoost in an SEO audit

What is NavBoost?

NavBoost is a query-to-document record of user clicks that Google uses as a ranking signal. The clearest one-line definition comes from DOJ-filed notes of a January 2025 call with Google engineer Pandu Nayak: NavBoost is “a measure of how frequently users (subset by location and device type) click on a particular document for a particular query,” a “traditional signal” that “uses most recent 13 months of data.”

The same notes describe its structure as “a QD table, a query-to-document lookup table, used in both directions containing counts/frequencies of user query activity by document.” In plain terms: for each query, Google knows which documents people clicked, and for each document, which queries led people to it.

It is one signal among many. The Nayak notes list it next to Q* (“Google’s measure of quality of a document”), RankEmbed (“one of Google’s primary LLM-trained signals”), PageRank, and twiddlers that “re-rank a set of already selected results.” The diagram shows where it sits.

Scroll the diagram sideways to see all of it.

What the DOJ case revealed about NavBoost

The antitrust case put NavBoost on the public record, with numbers. The plaintiffs’ post-trial brief, dated February 9, 2024, makes three statements worth knowing, each citing trial exhibits and testimony:

1. NavBoost is “one of the most powerful ranking components” at Google, and “memorizes all clicks for all queries received in the prior 13 months.”
2. “It would take Bing over 17 years to generate a similar amount of data.” That was the plaintiffs’ argument about scale as a competitive moat.
3. “Google then uses a system called Glue to help rank whole page results,” and “Glue also relies on enormous amounts of user data.”

The later Nayak call notes add an anecdote about why raw clicks are not used naively. An early Google signal measured how many times a link was shown against how many times it was clicked, and “it was found that this measure was biased by link position.” The result in position 1 gets clicked more because it is in position 1. The notes describe a modification that accounted for position, although parts of that passage are redacted.

That detail matters for how you read everything else. NavBoost is not a raw click-through-rate leaderboard. Google has known for a long time that position inflates clicks, and it corrects for it.

The evidence comes from three kinds of source, and they carry different weight:

None of the three gives a formula. Together they establish that NavBoost exists, that it uses clicks at scale, and what a click record contains. Bing, for comparison, simply lists user engagement as a ranking parameter in public, as our Bing SEO guide shows.

The click attributes in the leaked documents

The 2024 leak shows what a NavBoost click record contains. The module QualityNavboostCrapsCrapsClickSignals is described as “Click / impression signals for craps,” and the related QualityNavboostCrapsCrapsData holds the same counts along with the slices they are aggregated by. SparkToro’s write-up of the leak tied goodClicks, badClicks and lastLongestClicks to NavBoost.

Google stores these attributes; the leak does not show how much each is weighted. Many carry no description at all, which is itself informative: the names are all we have for goodClicks and badClicks.

Scroll the diagram sideways to see all of it.

Three descriptions stand out:

• country: “The two-letter uppercase country slice of the CrapsData. Examples: ‘US’, ‘FR’, ‘BR’.” Clicks are counted per market, which matches the Nayak notes on location slices.
• voterTokenCount: “The number of distinct voter tokens (a lower bound on the number of distinct users that contributed to the entry, used for privacy-related filtering).”
• unicornClicks: “The subset of clicks that are associated with an event from a Unicorn user.” The documentation does not say what a Unicorn user is. Treat any explanation you read elsewhere as a guess.

There is also a locale-level multiplier in a separate module, CountryMetroNBFeature.navboost, described as “the multiplier to apply to the result for this locale & query,” with a note that it “is for serving purposes only.” That is as close as the leak gets to showing NavBoost changing a result, and it still gives no number.

What is a last longest click?

A last longest click is, in the leak’s words, a click that was “last and longest in related user queries.” The description sits on QualityNavboostCrapsCrapsData.lastLongestClicks. Read literally, it is the result a searcher clicked last, and stayed on longest, while working through a set of related queries. In other words, the result that ended the search.

That is a better proxy for satisfaction than a click alone. A searcher who clicks your result, returns quickly and clicks a competitor has told Google something about both pages. The diagram illustrates the idea; it is our interpretation of the field names, not a documented algorithm.

Scroll the diagram sideways to see all of it.

The practical goal follows directly: be the result that ends the search. That is a content question, not a click question. It means answering the actual intent fully enough that the searcher has no reason to go back, and showing the first-hand experience that makes a reader trust the answer, which is the ground our E-E-A-T guide covers. Our guide to search intent covers how to read what a query is really asking.

Why fake clicks are a bad bet

Click manipulation services sell the idea that more clicks mean higher rankings. The evidence points the other way. Three things in the primary sources make it a poor investment:

• Filtering exists. voterTokenCount records a lower bound on distinct users behind an entry, “used for privacy-related filtering.” unscaledIpPriorBadFraction is “used to assign a prior based on IP address.” Signals like these suggest clicks are weighed for how many real, distinct users produced them and where they came from.
• Position is corrected for. The Nayak notes describe Google finding that click rates were “biased by link position” and building a correction. A crude CTR boost is exactly what that kind of correction is designed to discount.
• The valuable click is the last one. Bots can click. They cannot make a real person stop searching. If lastLongestClicks means what its description says, fake traffic does not produce it.

None of this proves any given manipulation will fail; the weights are unknown. It does mean you would be paying to fight a system designed with your tactic in mind.

What NavBoost means for titles and snippets

Titles and snippets decide the first click, and pages decide whether it counts. Four habits follow from what NavBoost records:

1. Promise exactly what the page delivers. A title that overpromises can win the click and lose the session. Google’s helpful content questions ask whether a title “avoid[s] exaggerating or being shocking in nature.” Our guide to SEO title length covers the mechanics.
2. Match the query’s wording where it is honest. The leak also stores a site-level titlematchScore, described as how well “titles are matching user queries.” We cover it in our Google leak explainer.
3. Answer early. A searcher who has to scroll past three paragraphs of preamble is more likely to go back. Put the answer in the first lines.
4. Write the snippet as a preview. The meta description should tell the searcher what they will find, so the click is informed. Our meta description examples show the difference.

Click data is sliced by country and device, so check performance by segment. A page that satisfies desktop users but is painful on mobile can be judged differently in each slice. Google’s Search Console Performance report can filter by device and country, and our CTR in SEO guide shows how to read the result.

Working with NavBoost in an SEO audit

We treat NavBoost as a reason to audit the match between query, title and page, not as a lever to pull. In practice that means three checks. First, compare each important page’s top queries with its title and first paragraph: does the page answer what people searched? Second, look for pages with high impressions and low CTR, where the snippet is failing, and pages with good CTR but weak engagement, where the page is failing. Third, segment both by device and country, because that is how the click records are stored.

Google stores these click attributes; the leak does not show how much they are weighted, and the trial record shows that NavBoost matters without giving a formula. That is enough to act on. If you want this query-to-page matching done across your whole site, it is part of our AI visibility audit, which also checks how the same pages perform in AI answers, where a satisfied click is even harder to earn.