An analyst opens a GA4 report at the end of the quarter and finds a page sitting at position three for a competitive query, with an engagement rate that looks alarming. Visitors land, skim, and leave. The instinct is immediate: fix the bounce, or the ranking will slide. So the team spends two weeks adding a video, a related-posts widget, and a sticky table of contents, watching the dashboard for a number that never quite moves the ranking either way.
That effort is aimed at the wrong target, and the reason is one of the most persistent confusions in search. Google has said, repeatedly and on the record, that it does not use Google Analytics metrics like bounce rate as a ranking signal. Court testimony and a large documentation leak, meanwhile, indicate that Google does use click data to reorder results. Both statements are true. Untangling them is the difference between chasing a dashboard and improving the thing the dashboard is a shadow of.
The short answer #
Google has stated it does not use Google Analytics bounce rate or on-site dwell time as a ranking factor. Separately, testimony from the United States antitrust trial and a 2024 leak of internal documentation indicate that Google does use its own click data from the search results to help reorder them, through a system reported as NavBoost. The engagement number in your analytics tool is not the signal. The behavior Google observes on its own results page is closer to it, and you influence that behavior by satisfying the query, not by decorating the page.
Two stories that seem to contradict each other #
For years, Google representatives have pushed back hard on the idea that engagement metrics drive rankings. John Mueller has called it a misconception that Google Analytics figures feed the algorithm. Gary Illyes has been blunter, stating that Google does not use analytics or bounce rate in search ranking and describing clicks in general as incredibly noisy and hard to clean up. Taken alone, that sounds like a closed case: stop worrying about engagement.
Then came the other story. During the United States Department of Justice antitrust trial against Google, a Google vice president of search testified under oath about a system referred to as NavBoost, describing it as one of the more important ranking components and confirming that it relies on click data. In May 2024, a large set of internal Google API documentation was accidentally published to a public code repository, exposing thousands of attributes, several of which describe click signals by name. Google acknowledged the documents were genuine but cautioned that they were, in its words, out of context, outdated, or incomplete. The reasonable reading is not that Google lied. It is that the two claims are about two different things.
What NavBoost appears to do #
Based on the trial testimony and the leaked documentation, NavBoost is best understood as a re-ranking layer rather than the way pages enter the running in the first place. Other systems assemble a pool of candidate pages for a query. NavBoost then uses aggregated click behavior to nudge that pool into a better order, according to how the testimony described it. It does not invent relevance from nothing. It refines an ordering that already exists.
The leaked attributes name a few click types worth knowing, with the caveat that names in leaked documentation are not a rulebook. One reads as a good click, a result the searcher stays on without bouncing back. Another reads as a bad click, the pattern where a user returns to the results almost immediately, a behavior known as pogo-sticking. A third, described in the documentation as a particularly strong signal, is the last and longest click in a session, the result a searcher lands on and stops, suggesting the search is finally over. Reporting on the testimony also described a rolling window of roughly a year over which this behavior is aggregated, and the use of Chrome data. Treat those specifics as informed indications, not settled fact, since Google itself flagged the material as incomplete.
Why both claims are true at once #
The reconciliation is straightforward. When Google says it does not use bounce rate, it is talking about the number in your analytics account, calculated by a third-party tag on the fraction of sites that install it, measured with definitions Google does not control. That figure is noisy, gameable, and absent for most of the web. When testimony describes NavBoost using clicks, it is talking about Google’s own first-party record of what searchers do on Google’s own results page, which Google measures directly and completely. A bounce in your analytics is on-site behavior. A bad click in NavBoost is search behavior, and the two are not the same event.
Pogo-sticking is the clearest example. A visitor who reads your answer, gets what they came for, and closes the tab is a bounce in your analytics and a satisfied searcher to Google. A visitor who clicks, recoils, and picks the next result is a short session on your page and a bad click on the results page. The analytics tool cannot tell those two apart. Google’s click stream can.
What this means for what you measure #
The practical move is to stop treating every engagement number as a lever and start reading each one for what it reflects. Some are diagnostic, useful for spotting a problem but not worth optimizing toward. Others are proxies for the search behavior that does matter, and they improve when the page improves, not when the metric is chased directly.
| Metric | Where it comes from | What it reflects | How to treat it |
|---|---|---|---|
| GA4 bounce or engagement rate | Your analytics tag, on a subset of visitors | On-page behavior, not search behavior | Diagnostic only; investigate outliers, do not optimize the number |
| Click-through rate in Search Console | Google’s own results page | How well the title and snippet match the query | A prompt to improve relevance of the title and snippet, not to manufacture clicks |
| Pogo-sticking, return to the results | Google’s click stream, visible to you only indirectly | Whether the click resolved the query | Reduce the gap between what the page promises and what it delivers |
| Dwell, the last and longest click | Google’s click stream, inferred | Whether the search journey ended on the page | Be the page that finishes the task rather than the one that sends people back |
The trap of trying to game the click #
Once people learn that clicks matter, the next idea is often to manufacture them, with click bots, incentivized searches, or misleading titles that win the click and lose the visitor. This tends to fail for the same reason Google gives when it downplays clicks: the signal is noisy, and Google spends real effort filtering manipulation out of it. A title that oversells earns the click and then earns the pogo-stick, which is the worst of both outcomes. Attention is spent getting a searcher to arrive disappointed, and the disappointment is exactly the part the system is designed to notice.
There is a quieter version of the same mistake: optimizing the analytics number directly. Adding an autoplay video to pad session time, or splitting one answer across five pages to lift pageviews, can move a dashboard while doing nothing for the searcher, and sometimes working against them. The metric improves. The thing the metric was standing in for gets worse.
What actually moves the signal #
If click behavior on the results page is what feeds re-ranking, then the work is to make the click resolve the query, and that work is ordinary. Pages that satisfy a search match the intent behind it, so the person who clicks finds what they came for. They answer the main question early and plainly, so the value is visible before anyone scrolls. Their title and snippet read as an honest preview, so expectation and delivery line up and there is no reason to bounce back. They stay fast and readable enough that a satisfied reader is not driven off by friction. None of that is a trick aimed at the algorithm. It is the same work that satisfies a person, which is the point: Google is trying to measure satisfaction, and the most durable way to score well on a proxy for satisfaction is to actually satisfy.
What we still do not know #
It is worth being clear about the edges of this. The trial testimony and the leak are the best public evidence available, but neither is a current, official specification, and Google explicitly warned that the leaked material was incomplete and out of date. How much weight click behavior carries, how it interacts with everything else, and how the system has changed since are not publicly settled. What is settled enough to act on is the shape of it: the number in your analytics dashboard is not the ranking signal, the searcher’s behavior on Google’s results is closer to it, and you reach that behavior only through the page itself. If you were about to spend two weeks lowering a bounce rate, spend them making the page the last click instead.
This reflects what is publicly known as of mid-2026, drawn from antitrust trial testimony and leaked internal documentation that Google described as incomplete. Google’s systems and public guidance change, and the specifics here may age.