Dwell time is the stretch between a searcher clicking your result and coming back to the results page. Google has said for years that it does not rank pages on a time-on-page number, and that is true as far as anyone can verify. But the US antitrust case against Google put on record that it does use aggregated click data per query and per page, through a system called Navboost. Both are true. Here is how they fit, why chasing seconds is the wrong goal, and how to read the engagement you can measure.
What Dwell Time Is, and the Metrics It Gets Confused With
The metric only exists on the search engine’s side. If the searcher never comes back, there is nothing to time, and that is usually the best outcome. Your analytics never sees the return trip, so no dashboard you own reports it.
Most arguments about it are really about a different metric:
| Metric | Who can see it | What it measures | Blind spot |
|---|---|---|---|
| Dwell time | The search engine only | Time from clicking a result to returning to that results page | Never published; undefined if the user doesn’t return |
| Pogo-sticking | The search engine only | A quick return followed by a click on a different result | An SEO term, not a metric Google documents |
| Average engagement time (GA4) | You | Time the page was in focus in the browser | Cannot tell whether the visitor went back to Google |
| Bounce rate (GA4) | You | Share of sessions that were not engaged | A satisfied five-second answer counts as a bounce |
| Organic CTR (Search Console) | You | Clicks divided by impressions for a query or page | Says nothing about what happened after the click |
What Google Has Said on the Record
Google’s public line has been blunt. In a February 2019 Reddit AMA, Gary Illyes wrote that “dwell time, CTR, whatever Fishkin’s new theory is, those are generally made up crap” (reported by Search Engine Roundtable). In December 2022, someone asked John Mueller why Search Console Insights doesn’t show how long people spend on a page. He replied on Twitter, “It doesn’t — that’s why there’s no number” (Search Engine Roundtable). In August 2023 he also mocked a LinkedIn tip to slow a site down to raise time on site.
Google’s own documentation is less absolute. Its How Search Works page on ranking says Google uses “aggregated and anonymized interaction data to assess whether search results are relevant to queries,” and turns that data into signals for its machine-learned systems. Its page on testing says live traffic experiments look at metrics “such as what people click on,” along with abandoned searches and how long it takes someone to click a result, before a change launches. So Google denies a time number, not interaction data.
What the Antitrust Case Revealed About Navboost
The DOJ case filled in the gap. One public exhibit, PXR0357, records a January 2025 call with Google’s Pandu Nayak. The notes describe Navboost as “a measure of how frequently users (subset by location and device type) click on a particular document for a particular query.” They call it a traditional signal that “uses most recent 13 months of data.” They add something most articles on this topic leave out: Google “avoids simply ‘predicting clicks’ because clicks are easily manipulated and are a poor proxy for enhancing user experience.” They also describe an early click signal that was biased by link position: the first result gets clicked more simply because it is first.
The 2024 leak of Google’s internal Content Warehouse API documentation named click attributes such as goodClicks, badClicks and lastLongestClicks. Google confirmed the documents were real, but it warned against reading them out of context. The leak showed the attributes exist, not how they are weighted or whether they are still used. Our Navboost explainer covers the system in depth.
What Google Measures and What It Doesn’t
Side by side, the statements and the court record don’t conflict as much as people claim:
- Measured: clicks counted per query and per page, split by location and device and gathered over about 13 months, plus aggregated interaction data used to judge relevance and to test changes.
- Measured, details unknown: some sense of whether a click was “good” or “bad,” going by the leaked attribute names. We don’t know how Google defines those, whether time is part of the definition, or how much they count.
- Not supported by any public evidence: a per-page time-on-page score, a threshold such as “keep visitors three minutes,” or anything Google reads from your analytics tag. Nothing in the record suggests GA4 data is a ranking input.
The practical reading: Google watches, in aggregate, which result for a query seems to end the search. Time may be one clue, but its own notes call raw clicks easy to game, so don’t expect a single time number to be used on its own. Treat engagement as an outcome, never a target you tune directly.
Why a Short Visit Can Be a Win and a Long One a Loss
Time only means something when you know what the searcher was trying to do. Checking a store’s opening hours takes seconds; comparing project management tools can take twenty minutes. Look at length and outcome together:
| Visit length | What the searcher does next | Most likely meaning |
|---|---|---|
| Short | Stops searching | Quick answer, satisfied (a good result for fact or tool queries) |
| Short | Returns and clicks a different result | Pogo-sticking: wrong intent, bad first screen or a misleading title |
| Long | Stops searching | Deep read or task completed (a good result for research queries) |
| Long | Returns and clicks a different result | Struggled: buried answer, slow page or incomplete coverage |
Only the right-hand column matters, and time can’t tell you which row you are in. Hiding the answer under a long intro can push a visit from row one to row four: longer, and worse.
Pogo-Sticking: The Pattern Worth Worrying About
The pattern that deserves attention is pogo-sticking: a searcher clicks you, comes back, and picks someone else. It usually has one of three causes:
- Intent mismatch. The query wants a comparison and you show a sales page, or it wants a quick definition and you give 3,000 words of history. Check the format the current top results use with the Search Intent Checker and our guide to search intent.
- A snippet that makes promises the page doesn’t keep. A title that says “free” on a page behind a paywall wins the click and loses the visitor. Test the wording in the SERP Snippet Preview and make sure the first screen delivers what the snippet promised.
- A bad first screen. Pop-ups, a slow main image or layout shift drive people back before they read anything. Our sibling guide on PageSpeed Insights covers how to diagnose this.
How to Read Engagement You Can’t See Directly
You can’t measure the real thing, but you can triangulate. Read three sources together, never one at a time:
- Search Console, per query and page: impressions, CTR and average position over 3 to 6 months. A page with normal CTR for its position that slowly loses ground deserves a closer look.
- GA4, organic landing pages only: engagement rate and average engagement time. GA4 counts a session as engaged if it lasts longer than 10 seconds, has a key event, or has 2 or more page views. It times engagement only while the page is in focus. The 10-second timer is adjustable in tag settings, so check it before comparing properties. See bounce rate vs engagement rate and GA4 user engagement for more.
- The live results page: what now ranks above you, which format it uses, and whether a SERP feature already answers the query.
SEO Rocket shortens this loop. With Search Console and GA4 connected, you can ask in chat for organic landing pages where position is slipping and engagement time sits well below the site’s norm. Rank tracking then shows the trend line, which matters more than any single day when the click data behind Navboost spans 13 months.
A Worked Example: Reading One Page
Here is a hypothetical. A guide ranks for a “how to” query. Search Console shows steady impressions and a CTR in line with its position, but average position has drifted down over two quarters. In GA4, organic sessions are mostly engaged with long engagement time. It looks healthy.
The results page tells a different story. The pages that overtook it put the steps in a numbered list near the top. This guide opens with four paragraphs of background. That fits row four of the table above: people spend time looking for the answer, and some give up and click a competitor. The fix is moving the steps to the first screen and cutting the preamble, not adding content.
After that change, average engagement time may well fall. That is fine: success is position and clicks over the next few months. If engagement time is your KPI, this fix will look like a regression.
What to Change on the Page
Every fix that reliably helps is a fix for the searcher, and longer visits are only a side effect:
- Answer in the first screen, then go deeper. Put the definition, the steps or the price range where nobody has to scroll for it.
- Match the format the query wants: a list, a table, a calculator or a comparison.
- Cover the next question so the visitor doesn’t go back to Google for it, and link to the natural follow-up.
- Fix the broken parts. Dead links, redirect chains and pages that show nothing until JavaScript runs cost visitors. SEO Rocket’s site audit runs a real crawler with JavaScript rendering and flags 404s, broken links and redirect chains before searchers hit them.
Tactics That Backfire
“Time on page matters” spreads faster than the nuance, so tactics that only inflate the clock keep coming back:
- Burying the answer under a long intro or a “jump to recipe” maze. This moves visits into the “struggled” row.
- Slowing the page down or adding autoplay video to pad the clock. Mueller’s reaction in 2023 says enough.
- Buying or automating clicks. Google’s own notes call clicks “easily manipulated,” which is why it says it avoids simply predicting them. You pay for traffic built to be discounted.
- Reporting GA4 engagement time as an SEO KPI. It tracks how heavy the content is more than whether searchers were satisfied. For better choices, see our sibling guide on website metrics.
The playbook behind SEO Rocket, proven across 1,000,000+ ranking pages, comes down to one thing: be the result that ends the search. That is why the article writer checks structure and completeness before a draft ships, not minutes on the page.
FAQ
Is dwell time a Google ranking factor?
Not as a published time score. Google has repeatedly said it doesn’t use time on page. Court exhibits confirm that Navboost uses aggregated clicks per query and per page over about 13 months. Time may feed into how Google judges a click, but there is no public evidence of how.
What is a good time on page?
There isn’t one. A good result for a quick-fact query can be seconds long, and a research query can take many minutes. Judge success by whether people stop searching, which shows up as stable or rising position and clicks.
Can I measure dwell time in GA4?
No. GA4 measures how long your page is in focus, but it can’t see a visitor going back to the results page. Use engagement time from organic landing pages as a rough proxy and read it next to Search Console data.
Does pogo-sticking hurt rankings?
Google hasn’t confirmed it as a named signal, but a quick switch to a competitor is exactly what aggregated click data could capture. Fix the intent match and first screen either way.
The Bottom Line
Time on the page is a useful idea and a poor target. Google doesn’t score pages on seconds, but it does learn, in aggregate, which results end the search. Earn that by answering faster and more completely than the page above you, and judge the work by rankings and clicks over months, not a timer.
