Google treats a mobile page as slow when the largest thing on the screen, usually a hero image or a headline block, takes longer than 2.5 seconds to appear. That measure is called Largest Contentful Paint, or LCP. In the Web Almanac 2025 performance chapter, HTTP Archive found that 62% of mobile pages come in under 2.5 seconds. The other 38% do not, and 13% of mobile sites sit in the band the report calls poor, over four seconds.
If nobody has ever measured your site, you have roughly one in three odds of being in the failing group. Here is how to check in ten minutes, and what a fix is worth.
The number that matters is measured on real phones, not your laptop
Your site probably feels fine when you check it. You are on a desktop, on office internet, with most of the page already cached. Visitors get none of that.
Mobile accounted for 51.51% of worldwide web page views in June 2026, against 47.12% for desktop, according to StatCounter Global Stats. StatCounter counts page requests rather than individual people, so read it as a rough split. More than half your views happen on a phone, often on a cell connection, often on a device a few years old.
Google collects timing data from real Chrome users and publishes it as the Chrome User Experience Report. Numbers measured on real visitors are called field data. Three of them make up Google’s Core Web Vitals.
Where mobile sites break down
Sites are getting better, slowly. HTTP Archive’s Web Almanac 2025 found 48% of mobile sites and 56% of desktop sites had good Core Web Vitals in 2025, up from 44% and 55% the year before.
The device gap is wider than that year over year gain. The same report puts good LCP at 74% of desktop sites versus 62% of mobile. A second measure, Interaction to Next Paint (INP), times how long a site takes to visibly respond after a tap. 77% of mobile sites have a good INP, against 97% of desktop. A tap that does nothing for a second reads to a customer as a broken button.
The usual culprit is JavaScript, the code that makes menus open and forms check themselves. HTTP Archive also tracks Total Blocking Time, a lab measure of how long a page is too busy running code to answer a tap. The median mobile figure rose to 1,916 milliseconds in 2025, up 58% from 1,209 milliseconds in 2024.
Being a big brand does not save you. In the same report, 51% of the 1,000 most popular mobile sites pass all three Core Web Vitals. A four page site with one plugin often beats one that has been collecting add-ons for six years.
What a tenth of a second was worth in a study of 30 million sessions
In 2020, Google, agency 55, and Deloitte published Milliseconds Make Millions, covering 37 European and US brand sites and 30.5 million user sessions. Mobile load times were monitored hourly for 30 days, analytics setups were audited first, and no site redesigned anything during the window.
A 0.1 second improvement across four speed metrics produced a 21.6% increase in users who got as far as the form submission page on lead-generation sites, plus a 7% increase in page views. Retail conversions rose 8.4% and travel rose 10.1%.
The gap between 3.4 and 3.3 seconds is invisible to you at your desk. One in five more people reached the form. That data was collected in October and November 2019.
The A/B test that showed cause, not correlation
Correlation has a hole in it. Fast sites tend to be well-run sites, and well-run sites convert better for reasons unrelated to load time.
Vodafone’s test with Google, published in 2021, closes it. A 50/50 A/B test ran roughly 100,000 clicks and 34,000 visits per day per variant, and the two versions were visually and functionally identical, differing only in speed optimization. A 31% improvement in LCP produced 8% more total sales, a 15% better lead-to-visit rate, and an 11% better cart-to-visit rate. Same design, same copy, same traffic, faster page, more sales.
The excuse the data kills
“That research is old.” It is. The Vodafone test is from 2021 and Milliseconds Make Millions is from 2020. The most repeated number in the subject is older still: in 2018, Google and SOASTA Research reported through Think with Google that as mobile page load time goes from one second to ten seconds, the probability of a visitor bouncing increases 123%. That came from a deep neural network trained on bounce and conversion data at a reported 90% prediction accuracy, plus an analysis of 11 million mobile landing pages on 4G. It is the oldest figure here, and it gets misquoted constantly, usually with the one-to-ten-second range stripped out.
We keep citing these three because nothing better has replaced them: they are large, their methods are written down, and no study of comparable scale has been published since. Citing a 2020 study with its date attached beats a fresher number somebody invented.
How to check your own site in ten minutes
-
Open PageSpeed Insights at pagespeed.web.dev, enter your homepage address, and read the Mobile tab. The top section shows real-user numbers from the Chrome User Experience Report, if Google has collected enough visits to your site to report them. Below it is a lab test, one simulated run on Google’s hardware. Field data is what your customers experienced; lab data is what a test phone experienced once. When they disagree, believe the field data.
-
Compare your mobile LCP against 2.5 seconds, the threshold behind the HTTP Archive numbers above. Over four seconds puts you in the poor band with 13% of mobile sites. Check your main service page too, not only the homepage.
-
Open Google Search Console and find the Core Web Vitals report. PageSpeed Insights tests one address at a time; Search Console sorts every address Google has data for into good, needs improvement, and poor, so you can see whether the trouble is one page or a whole template.
Plenty of local business sites do not get enough traffic for Google to report field data, so you will see lab results only. Treat that score as a hint.
The fixes, ranked by effort
Most of this is not owner-fixable, and anyone who says otherwise is selling a plugin.
Two things are yours. Stop uploading full-resolution photos: a 4MB image off a camera or a stock site is one of the most common causes of a slow LCP, since the largest element on a page is usually an image. Resize before upload and the page gets faster with no code touched. Then take inventory of your third-party scripts: chat widgets, popup tools, heat maps, duplicate analytics tags, social feeds. Each loads code from somebody else’s server. Delete what you do not use.
The next tier needs a developer: cutting unused JavaScript and CSS that a theme or page builder loads on every page, deferring scripts so they stop blocking the first paint, adding caching, and serving files from a content delivery network closer to your visitors.
The last tier is rebuild territory: a theme that loads six sliders, a stack of plugins that each add their own scripts, hosting sized for a hobby blog. In an August 2025 Clutch survey of 406 US small business owners, 90% planned to invest in their website within twelve months, and speed and performance ranked second at 53%, behind a full redesign or visual refresh at 58%.
When speed work is the wrong answer
Speed multiplies whatever your site already does. A site that converts nobody will convert nobody faster.
We talk people out of performance work fairly often. Optimizing a service page that gets eleven sessions a month is money spent on the wrong page, so fix the page people land on first. A structure problem is the other common case. If a visitor cannot tell what you do in a few seconds, or meets a contact form with nine required fields, load time is not your bottleneck. The Vodafone test held design constant on purpose, and it says nothing about a fast bad page beating a slow good one.
Then there is the case we see most. You buy better hosting, compress the images, strip unused plugins, and still land above four seconds, because the theme underneath ships a megabyte of JavaScript on every page. Speed work has a ceiling, and the foundation sets it. Owners pay for three rounds and reach the cost of a rebuild anyway, on a site they still do not like.
Built properly the first time, speed stops being a project and becomes a property of the build: images sized right, few outside scripts, a template that loads only what the page needs. That is website design and development handled well, and it is the difference between stewarding a budget and paying twice for the same problem.
Run your homepage through PageSpeed Insights on the Mobile tab and write down the LCP. Under 2.5 seconds and you can leave speed alone for a year. Between 2.5 and four, start with the two fixes that are yours: image sizes and third-party scripts. Over four seconds, the question is whether the foundation can carry a fix at all. We answer that one at UpLift Digital for the price of a phone call.