If you’ve spent any time reading official Google statements, you’ve collected a stack of denials. Clicks aren’t a ranking factor. Bounce rate isn’t used. Dwell time isn’t measured. Google Analytics data never touches the algorithm. Each of these has been said, on the record, by a named Google representative, more than once. And then in 2023 a Google engineer sat in a federal courtroom and described, under oath, a system that ranks pages using clicks. A year later an internal API leak named the exact click signals it stores. So which is it?
The honest answer is that both are true, and the gap between them is the single most useful thing you can understand about how Google actually ranks. Once you see how Google’s public denials are worded, you stop reading them as contradictions and start reading them as carefully scoped technical statements that quietly confirm the opposite of what they appear to say. This post walks through the major denials one by one, lines each up against what the 2023 antitrust trial and 2024 leak revealed, and gives you a framework for reading any future Google statement so you’re never caught flat-footed by the next “we don’t use that” again.
Key takeaways
- The denials are technically accurate, not lies. Google reps deny using a specific tool or metric (your Google Analytics bounce rate, the literal “CTR” number), and they’re telling the truth. What they leave unsaid is that Google’s own ranking systems measure the underlying behavior through entirely different instrumentation.
- The 2023 DOJ trial and 2024 API leak are the Rosetta Stone. Sworn testimony confirmed NavBoost ranks with click data, and the leaked documentation named the signals – goodClicks, badClicks, lastLongestClick. These aren’t theories; they’re the source code’s own field names and a federal witness’s own words.
- Reading a denial correctly is a skill. Three questions – is this a tool or a behavior, a public spokesperson or a systems description, a “we don’t do that now” or a “we have never” – separate what Google is actually ruling out from what it’s letting you assume.
Why a Google denial is almost never a lie
It’s tempting to treat the contradiction as Google being dishonest. It isn’t, and assuming so will lead you to bad decisions. Google’s search liaisons are careful people answering narrow questions in public, and they answer the exact question asked – no more. The trick is that the question is almost always about a tool or a named metric, while the thing that actually affects your rankings is a behavior measured somewhere you can’t see.
Here’s the pattern in its purest form. Someone asks, “Does Google use bounce rate as a ranking factor?” A Google rep says no. That’s true – the bounce rate sitting in your Google Analytics dashboard, calculated by Google Analytics, is not piped into the ranking algorithm. But the behavior bounce rate approximates – a user clicking your result, not finding what they wanted, and returning to the search page to pick a different result – is exactly what Google’s systems record, under a different name, from a different data source. The denial is true and the behavior matters. Both things hold at once.
You’ll see this structure repeat across every signal below. Keep it in mind: Google denies the instrument, not the measurement.
There’s a second reason the denials are worded so carefully, and it’s worth understanding because it explains why Google will never stop denying the tools. If a Google rep ever said plainly “yes, we rank pages on how users click and engage,” they’d be handing every spammer on the internet a target. The entire incentive of public Google communication is to discourage manipulation – so the messaging consistently downplays anything an attacker could exploit. That’s not deception; it’s defensive vagueness. But it means the public statements are optimized for a goal other than informing you, and you have to read them with that in mind. The systems descriptions given under oath have no such incentive, which is exactly why they’re more revealing.
Denial #1: “We don’t use bounce rate”
This is the most repeated denial in Google’s history. Matt Cutts addressed it in 2010. Gary Illyes confirmed it again in 2015. John Mueller repeated it in 2022. Three separate reps, twelve years apart, all saying the same thing: bounce rate is not a ranking factor. And they’re right, in the specific sense they mean it. Google does not reach into your Google Analytics account and pull the bounce rate metric, and it would be architecturally strange if it did – Google can’t trust a number any site owner can configure, filter, or fake.
But the 2023 DOJ trial and the 2024 leak confirmed that NavBoost records every click, every return-to-SERP, and every dwell duration. A user who clicks your result and bounces straight back to Google to click a competitor is producing a signal Google absolutely tracks – it just isn’t your Analytics bounce rate, and it isn’t measured by Analytics at all. We’ve written the full breakdown of this specific gap in does bounce rate affect SEO rankings, because it’s the cleanest example of the whole denial pattern. If you only read one companion piece to this post, read that one.
The takeaway: when Google says “we don’t use bounce rate,” hear “we don’t use the Google Analytics metric called bounce rate.” Both are true. Neither means user behavior on your result is invisible to ranking.
Denial #2: “We don’t use Google Analytics data”
Closely related, and just as widely misunderstood. Google reps have stated plainly that Google Analytics data is not used for ranking. This one is almost certainly the most literally true denial of the set – mixing a free analytics product’s data into core ranking would create an obvious incentive problem and a privacy minefield, and Google has been consistent about it for years.
Here’s the misdirection, though. People hear “Google doesn’t use Analytics data” and conclude “Google doesn’t have my engagement data.” Those are completely different claims. Google doesn’t need your Analytics account to know how people behave on your result, because it has a far better vantage point: the click itself happens on Google’s own search results page, and a huge share of users browse with Chrome, which gives Google a first-party view of what happens after the click. The 2024 leak surfaced fields tied to Chrome-based signals – we unpacked what chromeInTotal and the related attributes actually showed in Chrome data as a ranking signal. Google ruled out one weak data source while sitting on a much stronger one. The lesson generalizes: when Google denies a data source, check whether it has a better source for the same information. It almost always does, and the denial quietly steers you away from noticing it.
Denial #3: “Clicks are not a ranking factor”
This is the big one, and the denial that aged the worst. For years the public line was some version of “clicks are too noisy to use directly” – the implication being that click-through rate is a vanity metric Google watches but doesn’t rank on. John Mueller in particular has framed clicks as noisy and easy to manipulate, which is a reasonable engineering concern.
Then came the 2023 antitrust trial. Under oath, a Google engineer described NavBoost, a system that has used click data to adjust rankings for the better part of two decades. Not as a curiosity. As a core component. The 2024 API leak then named the exact signals: goodClicks, badClicks, and lastLongestClick – the field names Google’s own systems use to distinguish a satisfied click from an unsatisfied one. We broke down what each of those three signals captures in goodClicks, badClicks, and lastLongestClick, and the architecture of the system that consumes them in what is NavBoost and how NavBoost uses your click data.
So how do you reconcile “clicks are too noisy to use” with “we built a click-based ranking system and ran it for eighteen years”? You reconcile it the same way as the others. “Too noisy to use directly” is true – Google doesn’t take a raw CTR number and bolt it onto a ranking score. It aggregates clicks across many users, filters the ones that look manipulative, weights them by satisfaction, and feeds the cleaned signal into NavBoost. The denial addressed the naive version (raw CTR as a direct factor) and let everyone assume it covered the sophisticated version (filtered, aggregated click satisfaction as a ranking input). It doesn’t.
Denial #4: “We don’t measure dwell time”
The fourth denial follows the same shape. Google has said it doesn’t use “dwell time” or “time on site” as a ranking factor, and once again the specific phrasing matters. There is no field called “dwell time” being read off a stopwatch. But lastLongestClick – one of the leaked signals – captures something functionally close: the click in a search session that led to the longest engagement before the user stopped searching. That’s a satisfaction proxy built directly from behavior duration. Whether you call it “dwell time” or not, Google is measuring how long the winning result held the user. We covered why this matters and how it compounds in why dwell time matters for SEO.
The same logic extends to repeat behavior. A user who returns to your site directly, without searching again, is producing one of the strongest satisfaction signals there is – and it’s measured, not denied. We walked through that mechanism in return visits as a user engagement signal.
The Rosetta Stone: the 2023 trial and the 2024 leak
Two events turned this from speculation into documented fact, and it’s worth being precise about what each one actually established, because they’re often blurred together.
The 2023 DOJ antitrust trial produced sworn testimony. A Google engineer, under oath and under penalty of perjury, described NavBoost as a system that uses click data in ranking. This is the strongest possible form of confirmation – not a leak that could be misread, not a patent that might never have shipped, but a Google employee describing the system in a federal courtroom with lawyers cross-examining every word. It established the existence and purpose of click-based ranking.
The 2024 API documentation leak produced field names. Thousands of attributes from Google’s internal Content Warehouse API surfaced publicly, including the click-signal fields – goodClicks, badClicks, lastLongestClick, and a web of related attributes. A leak alone is weaker evidence than testimony; documentation can describe things that were deprecated, experimental, or never used. But the leak didn’t stand alone. It corroborated the testimony, naming the precise mechanics of the system a Google engineer had already confirmed existed. Testimony gave you the what; the leak gave you the how. Together they’re as close to ground truth as anyone outside Google will get.
This is also why we lean on the human side of the system to interpret it. The behaviors NavBoost rewards line up with what Google’s human quality raters are trained to mark as satisfying – we connected those two worlds in how the Quality Rater Guidelines shape what NavBoost rewards. The denials describe the inputs Google won’t use; the rater guidelines describe the outcome Google is optimizing toward.
The one denial Google has never made
Step back and look at the whole collection of statements and a striking absence appears. Google has denied bounce rate. It has denied Google Analytics data. It has denied raw CTR as a direct factor. It has denied dwell time as a named metric. That’s four specific instruments ruled out. But across more than a decade of public communication, Google has never once said that user satisfaction is irrelevant to ranking. It has never said “we don’t care whether people are happy with the results they click.” It has never claimed it has no way of knowing whether a result satisfied the searcher.
That absence is the loudest signal of all. If user behavior genuinely didn’t matter, the simplest, most manipulation-proof thing Google could say would be “stop optimizing for clicks, we don’t measure them, it does nothing.” It would end an entire industry of speculation in one sentence. Google has had a dozen years and countless opportunities to say it. It never has – because it can’t, truthfully. The denials are a map of exactly where the boundary sits: Google will tell you which tools it ignores, and it will never tell you the behavior those tools approximate doesn’t count. Once you notice the shape of what’s denied versus what’s conspicuously left alone, the official messaging stops being confusing and starts being almost helpful.
A framework for reading any Google denial
You don’t need to memorize the history. You need three questions you can ask of any statement a Google rep makes, and they’ll tell you what’s actually being ruled out. Run every “we don’t use X” through these:
1. Is this a tool, or a behavior?
If the denial names a tool or a specific metric – “Google Analytics,” “bounce rate,” “the CTR number” – it’s almost certainly true and almost certainly narrow. Google can deny the instrument while still measuring the behavior through its own instrumentation. If the denial somehow claimed a whole behavior was invisible – “we have no idea whether users are satisfied with a result” – that would be the remarkable claim, and Google never makes it. They deny tools. They don’t deny that user satisfaction matters.
2. Is this a public spokesperson, or a systems description?
A search liaison answering a question on social media is performing a different job than an engineer describing architecture under oath. The spokesperson’s job is to prevent manipulation and avoid revealing mechanics; the testimony’s job is to be accurate or face perjury. When public messaging and sworn testimony disagree, weight the testimony. That’s not cynicism – it’s just understanding the different incentives and constraints behind each statement.
3. Is this “we don’t now,” or “we never have”?
Watch the tense and scope. “We don’t use that” describes the present and can quietly change. “We have never used clicks in ranking” would be a falsifiable claim about the whole history – and notably, Google doesn’t make that one, because NavBoost has been running for the better part of twenty years. The narrower the time scope of a denial, the less it actually forecloses.
Run any future Google statement through those three and the contradictions dissolve. You’ll find that Google has been remarkably consistent: it denies the tools, protects the mechanics, and never actually claims that what your users do after they click is invisible to ranking.
What this means for your site
The practical conclusion is the one that underpins everything we do. If Google’s ranking systems measure click satisfaction, return visits, and the behaviors that bounce rate and dwell time only approximate, then your rankings are downstream of how real users engage with your result – not the metrics in your Analytics dashboard, but the satisfaction signals Google collects at the source. Two pages with identical content and identical backlinks can rank differently because users behave differently on them, and that behavior is a signal you can influence.
Consider what that looks like in practice. Picture two competing pages sitting next to each other at positions 4 and 5 for the same query, with comparable content and comparable links. Over a few months, the page at position 5 starts earning a noticeably higher share of clicks for that query, those clicks tend to be the last one searchers make before they stop searching, and a growing slice of those visitors come back to the site directly later. None of that shows up in either site’s Google Analytics in a way Google reads. But every piece of it is exactly what NavBoost records at the source – and it’s the kind of behavioral gap that precedes the lower-positioned page climbing past the one above it. The denials say none of these metrics are “ranking factors.” The behavior underneath them is what moved the rankings anyway. That’s the whole reconciliation in one scenario.
This is exactly the foundation User Signal Amplification is built on, and it’s why the safety question has a real answer rather than a hand-wave. We make the complete case – including why Google’s architecture filters suspicious signals instead of penalizing the sites they point at – in is User Signal Amplification safe. The denials don’t undermine that case. Read correctly, they confirm it.
Frequently asked questions
Q: Is Google lying when it says clicks aren’t a ranking factor?
A: No, and treating it as a lie will mislead you. Google’s denials are narrowly scoped to specific tools or metrics – your Google Analytics bounce rate, the raw CTR number used directly. Those statements are technically accurate. What the denials leave unstated is that Google’s own ranking systems, confirmed by the 2023 DOJ trial and named in the 2024 API leak, measure the underlying user behavior through different instrumentation. The denial is true; the behavior still affects rankings.
Q: What exactly did the 2023 antitrust trial and 2024 leak confirm?
A: The 2023 DOJ trial produced sworn testimony from a Google engineer describing NavBoost, a system that uses click data to adjust rankings and has done so for nearly two decades. The 2024 API documentation leak then surfaced the internal field names for the click signals – goodClicks, badClicks, and lastLongestClick – corroborating the testimony with the system’s own mechanics. Testimony established that click-based ranking exists; the leak named how it works.
Q: How do I tell what a Google statement is really ruling out?
A: Ask three questions. Is it denying a tool or metric (narrow, usually true) or a whole behavior (a claim Google never actually makes)? Is it a public spokesperson managing messaging or a systems description given under oath (weight the testimony)? And is it “we don’t do that now” or “we never have” (the narrower the scope, the less it forecloses)? Those three questions separate what’s genuinely ruled out from what you’re being left to assume.