You have probably read that a high bounce rate hurts your rankings, or that Google watches how long people stay on your page before bouncing back to the search results. That claim shows up in a lot of SEO guides, and it usually gets repeated as a settled fact in either direction: either Google obviously tracks this, or Google has said it never does. Neither one is quite right.
Here is the honest verdict. Google has explicitly denied using Google Analytics bounce rate to rank websites, while its own documentation and recent court records support the idea that it uses aggregated search interactions, including a historical signal built from how long someone stayed on a page before returning to search. That does not add up to a standalone dwell time ranking factor, a pogo sticking Google penalty, or any kind of bounce rate SEO target you can chase. Evidence grade: mixed. The rest of this piece walks through why.
What dwell time, bounce rate, and pogo-sticking actually measure
Part of the confusion here is that these three terms get used as if they mean the same thing. They do not.
Dwell time is the stretch of time between clicking a search result and going back to the search results page. It starts with a search click and it ends with a return to search, not with the end of an analytics session. Someone can click around your site for a while before going back to Google, and that whole stretch counts. If nobody ever goes back to search, there is no completed dwell time interval to measure at all, which means a missing return should never be read as proof that someone was satisfied or dissatisfied.
Bounce rate means something different depending on which version of Google Analytics you are reading about. In GA4, bounce rate is the share of sessions that were not "engaged sessions." Google counts a session as engaged if it meets any one of three conditions: it lasted longer than roughly 10 seconds, it had a key event, or it included at least two page views. Those are alternatives, not a checklist you need to clear all three of. So bounce rate in GA4 is really just 100 percent minus your engagement rate, and it tells you nothing about whether the visitor went back to a search engine afterward.
Universal Analytics measured bounce rate differently: the share of sessions where someone viewed only one page and sent only one request to the Analytics server. Google's own documentation gives a useful example here: someone could read your homepage for several minutes and then leave without triggering a second request, and that visit still counts as a bounce under the old model. A bounced UA session is recorded with a reported duration of zero seconds, which is a quirk of the measurement, not proof the person read nothing.
Pogo-sticking describes a behavior, not a percentage in your analytics dashboard: someone moves back and forth between the search results and the pages they click into, often with short visits in between. Leaving a page is not the same as returning to search. Returning to search once is not the same as a repeated back-and-forth pattern. And comparing a few useful sources before picking one does not mean any of them failed.
| Concept | What it describes | What it does not establish |
|---|---|---|
| Dwell time | Time from a search click to a return to the results page | Active reading, satisfaction, or a disclosed Google weight |
| GA4 bounce rate | Share of sessions that miss the engaged-session criteria | Whether the visitor returned to search, or content quality on its own |
| UA bounce rate | Single-page, single-request sessions under the older model | That no reading happened |
| Pogo-sticking | Back-and-forth movement between results and destination pages | An automatic penalty, or a single identifiable motive |
Why these behaviors became alleged ranking factors
The term "dwell time" entered SEO conversation through a 2011 discussion attributed to Duane Forrester, who worked on Bing at the time. He described a rough rule of thumb: a visit of a minute or two was a good sign, a visit of only a couple of seconds was a weaker one, and this was one signal among several that Bing paid attention to. Over time, SEO writers generalized that Bing comment into a claim about Google, even though it came from a different search engine describing its own practices.
The deeper story underneath that is a plausible idea that got stretched into a mechanism. It makes intuitive sense that people spend longer with content they find useful. The trouble is that "people who like a page tend to stay longer" does not automatically prove that "staying longer makes Google rank you higher." A duration is, at best, an imperfect proxy for satisfaction, and watching that proxy move alongside rankings tells you nothing about how, or whether, a search engine actually used it.
Older analytics terminology made this worse. Under Universal Analytics, a reader could spend several focused minutes on one article and still register as a bounce, because bounce meant "one page, one server request," not "left quickly." Under GA4, that same reading pattern would usually count as an engaged session instead. A slogan like "bounces are bad for SEO" can quietly change meaning just because the underlying analytics definition changed, which is one reason this topic keeps resurfacing with conflicting advice attached.
A widely cited 2021 evidence review from Search Engine Journal concluded that dwell time was "definitely not a ranking factor" and that post-search behavior on a page did not factor into Google's rankings at all. That review was careful and its warning about correlation not proving causation still holds up. But it was also written before the court testimony and engineering records discussed next, and its blanket exclusion of post-click behavior turned out to be more categorical than the later record supports.
The evidence that Google uses post-click interaction signals
Start with what Google says about itself. Google's public explanation of how ranking works states plainly that it uses "aggregated and anonymized interaction data to assess whether search results are relevant to queries," and that it transforms that data into signals that help its machine-learned systems estimate relevance. That is a real, named acknowledgment that interaction data plays a role. It is not an admission that every click and return is tracked, and it is not a confirmation of a standalone dwell time score.
The more specific evidence comes from the US Department of Justice's antitrust case against Google. In testimony from September 20, 2023, former Google distinguished software engineer Eric Lehman discussed a system called Navboost. Asked whether it trains on roughly 13 months of user data, he agreed, then clarified something important: "Navboost is not a machine learning system. It's just a big table," built from aggregated counts of interactions tied to particular queries and particular documents. That is a search-side mechanism built from aggregated click and interaction records, which is a very different thing from a publisher's sitewide Analytics percentage.
A January 31, 2025 engineer-call record involving Google's Pandu Nayak describes Navboost the same way: a traditional signal based on interactions with a specific document for a specific query, broken down further by location and device, drawing on roughly the most recent 13 months of data. That window describes how the underlying data is aggregated. It is not a waiting period before a new page can rank, and it does not mean Google only updates this signal once a year.
The most direct corrective to the older "Google never looks at this" framing comes from a February 18, 2025 engineer-call record involving Google's Hyung-Jin Kim. Discussing fundamental ranking signals, the record describes clicks as covering, "historically, how long a user stayed at a particular linked page before bouncing back to the SERP." That word "historically" matters and should not be dropped: it is a description of how a signal was built over time, not a declared, currently fixed formula with disclosed weights. The same record notes that naming a system component is not enough, on its own, to recover the actual thresholds or curves behind it.
Put together, this is real, specific, dated evidence that something resembling post-click duration has figured into Google's engineering history. It is not a public specification of a dwell time ranking factor you could reverse-engineer and optimize toward.
What Google's denials do and don't rule out
Set that evidence next to what Google has said in public, because the two are easy to blur together.
In reporting published June 16, 2020, Google's John Mueller addressed the Analytics side of this directly: "I think there's a bit of misconception here that we're looking at things like the analytics bounce rate when it comes to ranking websites, and that's definitely not the case." That is a specific, attributed denial about one thing: the bounce rate percentage your Google Analytics dashboard shows you. The Navboost material discussed above concerns an internal search-side interaction table, not an ingestion of your Analytics property's reported bounce rate. Those are two separate claims, and the court records do not contradict Mueller's 2020 statement about Analytics specifically.
Mueller also addressed pogo-sticking directly, in a report published July 16, 2018: "We try not to use signals like that when it comes to search." He explained that people bounce back and forth for all kinds of reasons that have nothing to do with content quality: comparing a few options, checking a detail, or simply misclicking. He said it was hard to turn that kind of behavior into a reliable per-page ranking factor, and he separated that from the idea of Google studying aggregate reactions across millions of queries when evaluating a broader algorithm change. That 2018 statement is older than the court records above, but the two things it addresses, an automatic penalty for one return visit versus an aggregated interaction signal built at scale, were never actually the same claim.
Why a correlation chart can't prove a ranking mechanism
A fair amount of the "dwell time matters" case online rests on correlation studies: pages that rank well also tend to show better engagement numbers, therefore engagement must be driving rankings. A frequently cited Moz analysis by Larry Kim looked at relationships between bounce rate, time on site, and rankings within one particular context, and it is worth taking seriously as a prompt to investigate your own pages. It is not proof of a mechanism, and even Kim's own piece acknowledges several other explanations for the pattern it found.
A few of those alternative explanations are worth naming plainly. Better rankings can change who visits and what they do, which means the causality could run backward from what the chart implies. A page's underlying quality, or how well it matches what someone actually searched for, could be driving both the engagement numbers and the ranking at the same time, without either one causing the other. A sitewide bounce rate mixes very different page types, traffic sources, and visitor intents together, while the search-side signal described in the court records concerns a specific query paired with a specific document, which is a much narrower comparison. And a simple change to how you define a key event in your analytics setup can move your bounce rate without anything about the page or its ranking changing at all.
It also helps to separate three different things that get lumped under "Google uses behavior data." There is using aggregate behavior to judge whether a broader algorithm change improved results overall. There is building signals or internal tables from interaction data. And there is applying one specific metric as a direct, page-level ranking input. Evidence for the first does not automatically prove the third, and the fact that Google has talked about the first for years does not mean it has confirmed the third.
The verdict: use engagement to improve usefulness, not chase a score
So where does that leave a marketing lead looking at their own GA4 reports? Treat a weak engagement pattern as a reason to go look at a page, not as proof of a specific ranking problem on its own. A visitor who finds an opening time in six seconds and leaves is not a content failure just because GA4 might log a bounce. A visitor who spends several minutes struggling to find a buried answer is not succeeding just because the duration looks long. Duration alone tells you very little without knowing what the page was supposed to do for that visitor.
The more useful move is to compare engagement within similar page types, similar traffic sources, and a consistent measurement setup, rather than chasing one universal target across your whole site. Put the actual answer near the top instead of making people scroll for it. Write headings that describe the question they answer, so someone scanning the page can tell quickly whether they are in the right place. Fix the gap when a page promises one thing and delivers another. None of that is about gaming a hidden score. It is about making the page do its job well enough that whatever Google's systems eventually infer from real user behavior reflects something true.
This is also where a content team's own reporting discipline matters more than another dashboard metric. Keeping search performance, on-site engagement, and actual business outcomes as separate, clearly labeled numbers, rather than one blended story, is what lets you tell the difference between "this page got better" and "we changed how we measure it." DeepSmith's AI Visibility module exists for a related but distinct question: not whether your GA4 bounce rate moved, but whether AI engines like ChatGPT and Gemini are actually citing your pages for the questions your buyers ask. That is a separate signal from anything in this piece, and it is worth tracking on its own terms rather than folding it into a dwell time debate it does not belong in.
What would move this verdict further would be a clear, current technical disclosure that ties a specifically defined duration or return-to-results measure to a disclosed ranking role, rather than a historical description in a redacted call record. Absent that, the most defensible position stays the same: post-click interaction data appears to matter somewhere in Google's systems, but no public evidence establishes a standalone dwell time score, a bounce rate SEO threshold, or an automatic pogo sticking Google penalty that you can optimize toward directly.
