There is no cliff at three seconds. The number everyone quotes is real, but it came from Google in 2016 and it is worded more carefully than the retelling: 53% of visits are likely to be abandoned if pages take longer than 3 seconds to load. That is a figure pulled from analytics across thousands of sites, not a stopwatch that empties your site at the three second mark.
Where the 53% number actually comes from
It comes from a post on the Google Ad Manager blog, “The need for mobile speed,” published September 8, 2016 by Alex Shellhammer and Juliette Neel.
The line reads: 53% of visits are likely to be abandoned if pages take longer than 3 seconds to load. The footnote names the data behind it. Google Analytics, 3.7 thousand mobile web sites worldwide, March 2016.
Two things in that post matter more than the headline. The sentence says visits are likely to be abandoned. It does not say what those visitors did next, or why. And the same post says the average load time for mobile sites is 19 seconds over 3G connections, footnoted to a Webpagetest.org sample of 11.8 thousand mobile homepage domains on a fast 3G connection, February 2016.
Nineteen seconds. That is the world the 53% was measured in. The pages on the slow side of that three second line were not sitting at 3.2 seconds. The average was more than six times the line.
The follow up research, and where the retelling got fuzzy
Google went back to this a year later. The AMP Project blog published “New Industry Benchmarks for Mobile Page Speed” on February 28, 2017, citing Google/SOASTA Research, 2017. Its finding: as page load time goes from one second to seven seconds, the probability of a mobile site visitor bouncing increases 113%.
That post repeats the 53% number as well, and you can watch the wording firm up on the way through. The 2016 sentence: 53% of visits are likely to be abandoned if pages take longer than 3 seconds to load. The 2017 sentence: 53% of mobile site visitors leave a page that takes longer than three seconds to load. Visits turned into visitors, and likely to be abandoned turned into leave. Same figure, harder claim, one year apart, both published by Google.
The 2017 post also puts the average time to fully load a mobile landing page at 22 seconds. That is a different measurement from the 19 seconds in the 2016 post, and the same picture of how slow the web was when these numbers were taken.
Neither sentence says half your visitors leave at the three second mark. Both describe a slope that gets worse the longer a page takes. If somebody quotes you an exact second count as a hard rule, or a speed score as the reason AI names you, they are repeating a headline, not measuring your site.
The target Google actually publishes now
Three seconds is not the current number anyway.
Google’s developer site web.dev, on its Largest Contentful Paint page, last updated September 4, 2025, says sites should strive to have a Largest Contentful Paint of 2.5 seconds or less. It says that should be measured at the 75th percentile of page loads, split across mobile and desktop.
Read the second half of that. The bar is not your average visit. Three quarters of the people who land on your page should see the main content inside 2.5 seconds, including the one on a single bar of signal in somebody’s basement.
That percentile is built from real visits, and a small local site often does not get enough of them for Google to report a number at all. Why PageSpeed Insights shows no real user data for your site covers what to do when that section comes up blank.
What a tenth of a second was worth to somebody else
Deloitte published a study called “Milliseconds Make Millions,” dated March 24, 2020. It analyzed mobile site data over a four week period from retail, travel, luxury and lead generation brands across Europe and the US.
After a 0.1 second improvement in load time, the study reported:
- Retail conversions up 8.4%, and average order value up 9.2%
- Travel conversions up 10.1%
- Lead generation information pages, bounce rate improved by 8.3%
A tenth of a second. That is the part that sinks the three second rule. If nothing happened until a page crossed some line, a tenth of a second on either side of it would move nothing. It moved something. Speed acts like a slope the whole way down, not a gate you pass once and forget.
What none of this tells you about your business
Now the honest part.
Not one of these studies measured a plumber, a roofer, an HVAC company, or an electrician in a US town. Deloitte’s brands were retail, travel, luxury and lead generation across Europe and the US. Google’s 2016 figure came from mobile web sites that opted into sharing benchmark data, which is all the footnote claims. An emergency call at 9pm on a Sunday is a different kind of visit than someone browsing shoes on a couch.
These are also patterns across big piles of sites. Slow pages tend to be heavy pages, and heavy pages tend to be built poorly in other ways as well. What is published is the link between speed and the outcome, not a test run on one business with everything else held still. They tell you which direction to walk. They do not tell you what one second is worth to your phone.
Here is the case where you should not buy anything. Run the test first. If three quarters of your real visits already load the main content in under 2.5 seconds, speed is not what is costing you work. A rebuild sold to you as a speed fix would be money spent on a problem you do not have. Go look at your phone number, your service area pages, whether anyone picks up, and how fast a web lead gets called back. Nobody can promise you a call count from a faster page, and anyone who does is guessing.
What to do next
Plain steps, in order:
- Put your own address into Google PageSpeed Insights. It is free and public, and anyone can run it on any site, including your competitors.
- Look for the field data section at the top before you look at the score. That section is built from real visitors. The score below it is a lab simulation.
- If field data shows Largest Contentful Paint above 2.5 seconds, you have something worth fixing.
- Open your site on your own phone, on cell service, well away from the office wifi. That is the test that matches your customer.
- Fix the biggest thing first rather than everything at once. What actually makes a website load faster, in order works through that list.
If the fixes keep stacking up on an old site, a rebuild is a separate decision and a bigger one, and fix or rebuild is a question to answer on paper first. What that involves is set out on the website build page.
No client results are published here, so the only site that can be pointed at in full is this one. It scored 100 on Google’s PageSpeed test, on desktop and on mobile, when it launched in August 2026. Scores move a few points from run to run, so go run that test yourself. That is the whole point of a public one.