Does Google Use Clicks as a Ranking Signal? Here Is the Definitive Answer.

,

Written by

Luca Tagliaferro

Add Lucatagliaferro.com on Google
Table of Contents
  1. Part 1: What Google said publicly for 20 years
  2. Part 2: the 2004 patent they filed while saying none of this
  3. Part 3: The second patent: where goodClicks and badClicks get named
  4. Part 4: The DOJ Trial
  5. Part 5: The May 2024 API leak
  6. Part 6: Assembling the full picture
    1. Step 1: Initial ranking
    2. Step 2: NavBoost re-ranking
    3. Step 3: The COEC model
    4. Step 4: The lastLongestClick as the ultimate signal
  7. Part 7: Why Google denied it
  8. Part 8: What this means for your website
    1. Your title tag and meta description are your first-impression CTR mechanism
    2. Above-the-fold content determines whether a user stays or bounces
    3. Comprehensive content earns lastLongestClicks
    4. Internal linking to your most-clicked pages compounds the signal
    5. Your brand search volume is the purest form of click signal
  9. The Conclusion
    1. Sources:

For nearly two decades, Google told the SEO community one thing about user clicks. Under oath in a federal courtroom, its own executives confirmed something quite different.

This article is a synthesis of four primary sources:

  • a 2004 Google patent
  • sworn DOJ antitrust testimony
  • 14,000-document internal API leak
  • and the specific language of Google’s own engineers

assembled into a single coherent picture of how click behaviour actually works as a ranking signal.

By the end of this piece, you will understand exactly what Google measures, how it measures it, what it calls each signal internally, and, most importantly, what this means for anyone trying to rank a website in 2026.

Part 1: What Google said publicly for 20 years

Before we get to what is actually true, it is worth establishing the official record of what Google told the SEO community.

Gary Illyes, Google Search Analyst, Reddit AMA, 2019:

“Dwell time, CTR, whatever Fishkin’s new theory is, those are generally made up crap. Search is much more simple than people think.”

Gary Illyes, on clicks as a ranking signal, repeated over multiple conferences: Clicks are described as a “very noisy signal” — too susceptible to manipulation to be used reliably.

Matt Cutts, former head of Google Webspam: Using clicks directly in ranking would be “a mistake” due to manipulation risk.

Matt Cutts, in person to an SEO practitioner: “Google’s organic algorithm does not use any Google Chrome data.”

John Mueller, Google Search Advocate, January 2022: Chrome data is not used in ranking — the only Chrome input is the CrUX report for Core Web Vitals.

John Mueller, 2008: Users hitting the back button on a site would not impact that site’s crawling, indexing or ranking.

This was the official position, consistently maintained for roughly twenty years across conferences, Reddit AMAs, podcasts, tweets, and personal conversations with SEO practitioners.

Part 2: the 2004 patent they filed while saying none of this

Here is where the story starts to fracture.

In September 2004, fifteen years before Gary Illyes called click theories “made up crap”, Google filed a patent titled “Systems and methods for correlating document topicality and popularity“ (US8595225B1). The inventors listed were Amit Singhal and Urs Hoelzle: two of the most influential engineers in Google’s history.

The patent describes a system that does the following:

  1. Tracks user interactions with search results: when a user selects a document from a results page, that selection is recorded and associated with the query
  2. Assigns popularity scores: based on the frequency and pattern of these interactions, each document receives a score relative to other documents for the same query
  3. Correlates popularity with topicality: the system maps documents to specific topics and calculates “per-topic popularity,” meaning the signal is query-specific, not a general site-wide metric
  4. Uses these scores as a ranking signal: the patent states explicitly that the popularity score “may be used as a ranking signal in raising, or lowering, the ranking of the document within the returned search results”
patents.google /patent/US8595225B1/en

You will notice the patent does not mention clicks.

This is deliberate and consequential, and it is the reason many SEO consultants initially missed it when looking for the NavBoost patent. The document discusses “user interactions,” “navigational patterns” and “document selections.” These are all clicks. You cannot have a user navigational pattern that does not involve a user clicking on something.

The patent was filed in 2004. NavBoost. the system it describes. went live in approximately 2005, as confirmed later under oath by Google’s own VP of Search.

So at the precise moment Google was publicly telling the industry that clicks were too noisy to use, its engineering team was filing the patent that formalised a click-based ranking system. That system then ran quietly in the background for the next eighteen years.

Part 3: The second patent: where goodClicks and badClicks get named

The 2004 patent establishes the architecture. A later patent, US10229166B1 — “Modifying search result ranking based on implicit user feedback“, goes much further. It names the specific signal types and defines them with mathematical precision.

The patent describes a ranking method built on the following framework:

Short clicks = bad signal. A user clicks a result and returns to the SERP quickly. The patent describes this as evidence that “the page didn’t satisfy their question.” Internally, per the API leak (more on that shortly), this becomes badClicks.

short click image - luca tagliaferro seo consultant

Long clicks = good signal. A user clicks a result and stays for an extended period before returning, or does not return at all. The patent describes this as evidence of a “good page.” Internally, this becomes goodClicks.

Last longest clicks = the strongest signal of all. This is the click that ends the search session. The user searched, clicked through various results, and settled on one page where they spent the most time and did not go back to try another result. The patent defines this as the clearest possible indicator that a page completely satisfied the user’s query. Internally, this is lastLongestClicks.

It goes without saying that also short clicks can be a good signal in some cases. When the user finds quickly what they want on a page, this results in a positive user experience even if the click was short. A user checking “how tall is the Eiffel Tower” needs three seconds to be satisfied.

A user researching “SEO strategy for e-commerce” might need fifteen minutes.

This is the reason why the patent also specifies something that distinguishes this from a simple “time on page” metric. Google weighs each click by query type, user type, device, language and country.

Google understands this distinction and calibrates the scoring accordingly. This is why the system is more sophisticated than a raw dwell time calculation, and why Google’s public denials of “time on site as a ranking factor” were technically accurate while being substantively misleading.

The patent adds one more layer of detail that matters enormously for content strategy.

It specifies that click data is cross-referenced with explicit feedback from human quality raters, the people who score pages for Google’s Search Quality Rater Guidelines. This means the goodClicks signal is being calibrated against human expert judgements about page quality. The two systems validate each other.

Part 4: The DOJ Trial

In 2023, the United States Department of Justice brought an antitrust case against Google. As part of the proceedings, Google’s own senior executives were required to testify under oath about how the search ranking system actually works.

Pandu Nayak, Google’s VP of Search, Day 24 of trial testimony:

When asked whether NavBoost dates back to 2005, Nayak replied:

“It’s somewhere in that range. It might even be before that.”

He confirmed that NavBoost is “one of Google’s strongest ranking signals”: not a supplementary adjustment, not a tiebreaker, but one of the most powerful factors of where a page ranks.

The trial documents described NavBoost as: “Learning from this user feedback is perhaps the central way that web ranking has improved for 15 years.”

Not one of many ways. Perhaps the central way.

Eric Lehman, Google executive, on how NavBoost scores documents:

“Navboost does kind of the natural thing, which is, in the face of that kind of uncertainty, you take gentler measures. So you might modify the score of a document but more mildly than if you had more data.”

Translation: the more clicks a page accumulates for a query, the stronger the NavBoost signal. Pages with limited click data get mild adjustments. Pages with deep click histories get powerful ones.

The DOJ’s argument about Google’s scale advantage: During the trial, it was argued that user click data, what people click on and interact with on search results, made up the majority of “Google’s scale advantage,” and that simply sharing these user interactions would allow competitors to “mimic Google’s ranking.”

This is a remarkable statement.

The DOJ was arguing, and Google did not substantively contradict, that the secret sauce of Google’s ranking superiority is not its algorithm but its accumulated store of human click behaviour. The algorithm is what other search engines have. The click data is what they cannot replicate.

The trial also confirmed NavBoost’s architecture: a rolling 13-month window of aggregated click data, sliced by country, device, language and query, continuously feeding a re-ranking system that sits on top of Google’s initial algorithmic output.

Part 5: The May 2024 API leak

Four months after the trial concluded, in May 2024, over 14,000 pages of internal Google Search API documentation were accidentally exposed on GitHub. Rand Fishkin of SparkToro and Mike King of iPullRank published detailed analyses (see sources below).

The leak confirmed the trial testimony and added technical detail the testimony had not provided. Specifically, it named the internal systems:

  • NavBoost: confirmed as the primary click-signal processing system, operating on the 13-month rolling window described in the trial
  • Glue: NavBoost’s companion system, which processes click signals for non-blue-link SERP features: Knowledge Panels, image carousels, video results, People Also Ask boxes. As Nayak explained in testimony, “Glue is just another name for NavBoost that includes all of the other features on the page.”
  • Craps: the internal module name for the system that processes “click and impression signals.” The specific metrics it tracks: goodClicks, badClicks, lastLongestClicks, unsquashedClicks, unsquashedLastLongestClicks, and unicornClicks (the last representing particularly high-quality engagement signals).

This is where the story produces its irony.

In 2019, Gary Illyes told the SEO community that click theories were “generally made up crap.” The internal name of the module processing those exact click signals is, literally, Craps.

Whether this was a deliberate joke embedded in the codebase or a coincidence is impossible to know. What is certain is that the “crap” Gary Illyes was publicly dismissing was, at that precise moment, running under that name in Google’s production ranking infrastructure.

The leak also confirmed Chrome browser data integration — something multiple Googlers had specifically denied. The chrome_trans_clicks attribute is documented in the API, collecting click data from Chrome users to feed into the ranking systems.

Part 6: Assembling the full picture

Let us now put all four sources together into a single coherent model of how click behaviour actually works in Google’s ranking system.

Step 1: Initial ranking

When you search for a query, Google’s primary algorithmic system (internally called Mustang) generates an initial ranking based on traditional signals: keyword relevance, PageRank, content quality, technical SEO factors. This produces a list of potentially relevant pages.

Step 2: NavBoost re-ranking

Before those results are served to you, NavBoost steps in. It consults its 13-month rolling database of aggregated click behaviour from other users who searched for the same or similar queries. It asks: historically, which of these pages have users found most satisfying for this query?

Pages that have accumulated strong goodClicks and lastLongestClicks for this query get a ranking boost. Pages that have accumulated badClicks — short visits followed by immediate returns to the SERP — get a ranking reduction. The adjustments are proportional to the volume of data: pages with millions of query-matched clicks get strong adjustments; pages with limited data get mild ones.

Critically, NavBoost slices this data.

A UK user on mobile in London searching “SEO consultant” gets NavBoost scores derived from other UK mobile users searching the same term, not from global desktop traffic. The signal is contextually calibrated.

Step 3: The COEC model

Researchers, including A.J. Kohn, have analysed the patent evidence to conclude that NavBoost likely operates on a COEC model: Clicks Over Expected Clicks. Google does not simply count raw clicks. It compares your actual click count to the expected click count for your ranking position.

A page ranking in position 4 that receives dramatically more clicks than the average position-4 result gets a strong positive signal, because users are actively choosing it over the results Google ranked above it. A page ranking in position 1 that receives fewer clicks than expected gets a negative signal — because users are scrolling past it to find something better.

This is why CTR optimisation through compelling title tags and meta descriptions has real ranking consequences — and why clickbait that drives clicks but delivers a poor experience is actively counterproductive. The badClicks penalty will undo any CTR gain within the 13-month window.

Step 4: The lastLongestClick as the ultimate signal

The lastLongestClick is the crown jewel of the system. When a user’s search session ends on your page, when they clicked your result, engaged with your content, and did not return to the SERP to try something else, Google records that as the strongest possible evidence that your page satisfied the query completely.

This is what SEO practitioners call “earning the last click.” It is the equivalent of a user saying, without words, “this was the answer I was looking for.” NavBoost accumulates these signals across millions of users over 13 months and uses them to reinforce your ranking for that query.

A page that answers a question completely, including related questions, the next logical question, and the context a user needs to take action, will systematically earn more lastLongestClicks than a page that answers partially and sends users back to the SERP. This is a direct ranking mechanism, described in a patent filed in 2004, confirmed under oath in 2023, and named in a leaked API document in 2024.

Part 7: Why Google denied it

The obvious question: if Google had been running this system since 2005, why did its representatives spend twenty years telling the industry it did not exist?

There is a plausible and a cynical explanation.

The plausible explanation is that click signals are genuinely noisy and manipulable, and Google did not want to create an incentive for widespread click fraud. The squashed vs unsquashed data in the API leak confirms Google maintains raw click data specifically to detect and filter artificial patterns. NavBoost’s 13-month window provides stability (manipulation has to be sustained across a very long time period to stick, which makes it expensive and unreliable to execute at scale).

By publicly downplaying clicks, Google may have been protecting the integrity of a system that works precisely because most people do not know how to game it.

The cynical explanation is that Google’s public communications about its algorithm have never been primarily informational. They have been strategic. The company has commercial interests in managing how SEOs, publishers, and regulators think about its ranking systems. The DOJ antitrust case itself was partly built on the argument that Google’s accumulated user data constitutes an insurmountable competitive moat. Admitting that click data is “perhaps the central way that web ranking has improved for 15 years” — which is what the trial documents say — would have strengthened every antitrust argument made against the company for two decades.

Both explanations can be true simultaneously.

Part 8: What this means for your website

Understanding this system has practical consequences that go well beyond the theoretical.

Your title tag and meta description are your first-impression CTR mechanism

They determine whether a user clicks your result at all. But a misleading title that earns clicks and then generates badClicks is worse than a modest CTR with a well-matched promise. The goal is not maximum CTR — it is maximum CTR from users whose intent genuinely matches your content.

Above-the-fold content determines whether a user stays or bounces

The badClicks signal is generated when a user clicks and immediately returns. This happens in the first few seconds. If your page takes three seconds to load, displays a cookie wall, or makes the user scroll through irrelevant content before reaching what they searched for, you are generating badClicks at scale. Each one is a ranking vote against you, accumulated in NavBoost’s 13-month memory.

Comprehensive content earns lastLongestClicks

The page that answers the question, then answers the next logical question, then provides the context a user needs to take action, will earn the last click in more sessions than a page that answers partially.

This is the structural logic behind long-form, comprehensive content outperforming thin pages on competitive queries, not because Google rewards word count, but because comprehensive pages earn more lastLongestClicks.

Internal linking to your most-clicked pages compounds the signal

The click distance patent (US8082246B2) establishes that proximity to highly-clicked, trusted pages passes authority. Structuring your internal links to connect your lower-traffic pages to your most-engaged content amplifies NavBoost signals across the site.

Your brand search volume is the purest form of click signal

When a user searches for your name directly, their journey starts and ends with you. Google’s own quality rater guidelines treat brand search as a proxy for entity authority.

Building brand recognition that drives direct name searches is, in a very literal sense, generating the cleanest possible lastLongestClick signals: ones that are essentially impossible to fake and impossible to misattribute.

The Conclusion

Here is the definitive answer to the question that opened this article.

Does Google use clicks as a ranking signal?

Yes. Confirmed by a 2004 patent filed by Google’s own engineers. Confirmed by sworn testimony from Google’s VP of Search in a federal court. Confirmed by 14,000 pages of internal API documentation. Confirmed by the DOJ’s own characterisation of click data as the majority of Google’s competitive advantage.

The system is called NavBoost. It has been running since approximately 2005. It tracks goodClicks, badClicks and lastLongestClicks for every query-document pair, aggregated across 13 months of user behaviour, sliced by device, country, language and user type. It re-ranks search results after the initial algorithmic pass. It is, by Google’s own court testimony, one of the search engine’s strongest ranking signals.

The internal module that processes these click signals is named Craps.

In 2019, Gary Illyes told the SEO community that dwell time and CTR theories were “generally made up crap.”

He was, in the most precisely literal sense imaginable, correct.

Sources:

Patents (Google Patents — primary sources)

  1. US8595225B1 — Systems and methods for correlating document topicality and popularity (2004, Amit Singhal & Urs Hoelzle) https://patents.google.com/patent/US8595225B1/en
  2. US10229166B1 — Modifying search result ranking based on implicit user feedback https://patents.google.com/patent/US10229166B1/en
  3. US8082246B2 — System and method for ranking search results using click distance https://patents.google.com/patent/US8082246B2/en

DOJ Antitrust Trial — primary sources

  1. DOJ Trial Exhibits — United States v. Google LLC (2020 Search case) https://www.justice.gov/atr/us-and-plaintiff-states-v-google-llc-2020-trial-exhibits
  2. DOJ Trial Exhibits — United States v. Google LLC (2023 Adtech case) https://www.justice.gov/atr/us-and-plaintiff-states-v-google-llc-2023-trial-exhibits
  3. Wikipedia summary — United States v. Google LLC (2020) https://en.wikipedia.org/wiki/United_States_v._Google_LLC_(2020)
  4. Wikipedia summary — United States v. Google LLC (2023) https://en.wikipedia.org/wiki/United_States_v._Google_LLC_(2023)

API Leak Analysis — primary analyst sources

  1. Rand Fishkin (SparkToro) — original leak analysis https://sparktoro.com (the specific post was published May 2024)
  2. Mike King (iPullRank) — technical API leak breakdown (published May 2024)

NavBoost Patent Analysis

  1. Roger Montti — Search Engine Journal — Could This Be the NavBoost Patent? https://www.searchenginejournal.com/navboost-patent/506975/
  2. Hobo-web — NavBoost: How User Interactions Rank Websites in Google https://www.hobo-web.co.uk/navboost-how-google-uses-large-scale-user-interaction-data-to-rank-websites/
  3. Hobo-web — The CRAP System: How Users Shape Google Results https://www.hobo-web.co.uk/craps/
  4. Hobo-web — United States DOJ v. Google LLC: Federal Antitrust Case Summary https://www.hobo-web.co.uk/united-states-v-google-llc-doj/
  5. Hueston — NavBoost: Google’s Hidden Click Ranking System Revealed https://hueston.co/digital-marketing/navboost-googles-hidden-click-ranking-system-revealed/
  6. SE Ranking — What Is Google’s NavBoost Algorithm and How to Optimize for It https://seranking.com/blog/navboost/
  7. Victorious — Navboost: What It Is and Why It Matters for SEO https://victorious.com/blog/navboost/

Click Signals Research

  1. Cyrus Shepard — Zyppy Signal — How Google Click Signals Drive SEO Rankings and AI Answers (March 2026) https://signal.zyppy.com/p/google-click-signals
  2. SEO by the Sea — Google Patents Click-Through User Feedback on Search Results to Improve Rankings https://www.seobythesea.com/2015/07/google-click-through-feedback-search-results/
  3. Stan Ventures — Google Search Ranking Signals Exposed: What DOJ Trial Docs Reveal https://www.stanventures.com/news/google-search-ranking-signals-exposed-what-doj-trial-docs-reveal-4300/
  4. GrowthMarketing.ai — 5 Ranking Lies Google Told — And How the DOJ Trial Proved Them Wrong https://growthmarketing.ai/ranking-lies-google-told/

Google Algorithm Leak Coverage

  1. Ascensor — Google Search Algorithm Leak: What You Need to Know https://www.ascensor.com/post/google-search-algorithm-leak–what-you-need-to-know
  2. Vizion Interactive — Everything You Need to Know About the Google Leak https://www.vizion.com/blog/everything-you-need-to-know-about-the-google-leak-misstatements-and-api-revelations/
  3. OuterBox — Google’s NavBoost Algorithm: A Highlight From the Leaked Google Search API Documents https://www.outerboxdesign.com/articles/googles-navboost-algorithm-highlight-from-leaked-google-search-api

DOJ Trial Coverage

  1. Search Engine Land — U.S. vs. Google Antitrust Trial: Everything You Need to Know https://searchengineland.com/google-search-antirust-trial-hearing-updates-431977
  2. Hobo-web — How Google Works in 2025: DOJ Antitrust Trial Disclosures https://www.hobo-web.co.uk/how-google-works/
  3. Tech-Insider — Google’s 90% Search Monopoly Faces DOJ Breakup (2026 update) https://tech-insider.org/google-antitrust-appeal-doj-search-monopoly-2026/

Previous Story

Next Story