- Home
- Speed study 2026
SME website speed study 2026
How fast are SME websites on a phone, and does the way a site is built make a difference? To find out, in this website speed study we tested 85 websites with Google PageSpeed Insights, the free tool anyone can use, and grouped them by the kind of web agency that built them. This page publishes the aggregated results, the method and the limits of the data. It names no third-party sites or companies.
On 28 October 2026 we tested 85 SME websites with Google PageSpeed Insights. Client sites built by hand-coding studios had a median mobile score of 91.5 and 57% scored 90 or more, while client sites built by template, WordPress and site-builder agencies had a median of 68.5 and 17% scored 90 or more. Slow loading of the main content, not layout shifts or heavy scripts, was the most common problem.
Illustration. Times are medians of Largest Contentful Paint (mobile, PageSpeed Insights) from our measurement of 85 SME websites on 28 Oct 2026: 30 client sites of hand-coding studios and 12 client sites of template and WordPress agencies. Read the study
Key findings
Hand-coding studios built faster client sites than template, WordPress and site-builder agencies, and the difference came almost entirely from how quickly the main content appeared on a phone.
- Most hand-coded client sites reached 90. 17 of 30 (57%) scored 90 or more on mobile, against 2 of 12 (17%) of the client sites built by template, WordPress and site-builder agencies.
- The main image was the bottleneck. Median mobile Largest Contentful Paint (LCP) was 2.95 seconds for the hand-coded client sites and 8.35 seconds for the client sites built by template, WordPress and site-builder agencies. Across all 85 pages, 35 took more than 4 seconds to show their largest element.
- A fast LCP meant a good score. All 25 pages with a mobile LCP of 2.5 seconds or less scored 90 or more.
- Layout shift and blocking scripts were rarely the problem. 79 of 85 pages had a layout shift of 0.1 or less, and 79 of 85 had 200 milliseconds of blocking time or less.
- Mobile is the hard test. 73 of 85 pages scored higher on desktop than on mobile. The median desktop score was 98, the median mobile score 84.
- Scores vary between runs. On the 13 sites we tested three times, the gap between the best and worst run ranged from 0 to 31 points, with a median of 4.
- Real-user data is rare. Only 14 of the 85 pages had field data from the Chrome User Experience Report. The 12 that passed Core Web Vitals had lab scores anywhere from 48 to 99.
- No WordPress page reached 90. Of the 8 pages we identified as WordPress from their page source, the best mobile score was 74.
How did we measure?
We tested every page with the public PageSpeed Insights tool at pagespeed.web.dev, using its default settings, and recorded the results from Google's saved reports.
- Tool: Google PageSpeed Insights with Lighthouse 13.5.0. The mobile test emulated a Moto G Power phone on a simulated slow 4G connection; the desktop test used PageSpeed Insights' desktop profile. Each analysis produced both.
- Date: 28 October 2026, between 00:44 and 02:42 Finnish time (EET).
- Runs: one run for every page, and three runs for a core set of 13 pages. For those 13, every figure is the median of the three runs.
- Metrics: the performance score, First Contentful Paint, Largest Contentful Paint, Total Blocking Time, Cumulative Layout Shift and Speed Index, on mobile and on desktop.
- Field data: the mobile Core Web Vitals assessment from the Chrome User Experience Report, for the tested address, covering the previous 28 days, where Google had enough data.
- Medians: where a group has an even number of sites, the median is the average of the two middle values, which is why some medians end in .5.
- Platform: identified from each page's source code and server headers, used only for the WordPress note above.
Every run has a permanent Google report link, and we keep all of them on file.
Which websites did we test?
We tried 89 web addresses and got a complete result for 85. Two addresses returned errors before testing and two could not be measured by PageSpeed Insights.
The pages come from 35 web agencies in Finland, elsewhere in Europe and North America, plus three live sites we built ourselves. We chose two kinds of agency. The first are studios that write their own code or build with lightweight static tools, most of which advertise speed. The second, our comparison group, are agencies that build on templates, WordPress or site builders. For each agency we tested its own website and up to three client sites linked from its portfolio or reference pages.
| Group | What it contains | Sites |
|---|---|---|
| A | Client sites built by hand-coding studios | 30 |
| B | Those studios' own websites | 22 |
| C | Portfolio sites from the same studios that were built on site builders or other platforms, and one demo site | 5 |
| D | Client sites built by template, WordPress and site-builder agencies | 12 |
| E | Those agencies' own websites | 13 |
| F | Live websites we built | 3 |
| All | 85 |
What were the results?
Groups A and B scored highest on mobile, and groups C, D and E lowest. The table shows the mobile results by group.
| Group | Sites | Median mobile score | Share scoring 90+ | Median mobile LCP | Share with LCP of 2.5 s or less |
|---|---|---|---|---|---|
| A: Client sites, hand-coding studios | 30 | 91.5 | 57% (17) | 2.95 s | 40% (12) |
| B: Hand-coding studios' own sites | 22 | 93.5 | 59% (13) | 2.95 s | 45% (10) |
| C: Builder and demo sites in those portfolios | 5 | 62 | 20% (1) | 10.3 s | 0% (0) |
| D: Client sites, template, WordPress and builder agencies | 12 | 68.5 | 17% (2) | 8.35 s | 17% (2) |
| E: Those agencies' own sites | 13 | 65 | 8% (1) | 7.2 s | 0% (0) |
| F: Sites we built | 3 | 90 | 67% (2) | 3.0 s | 33% (1) |
| All sites | 85 | 84 | 42% (36) | 3.7 s | 29% (25) |
Desktop scores were high in almost every group, and layout shift and blocking time were rarely a problem:
| Group | Median desktop score | Layout shift (CLS) of 0.1 or less | Blocking time (TBT) of 200 ms or less |
|---|---|---|---|
| A | 99 | 30 of 30 | 28 of 30 |
| B | 99.5 | 22 of 22 | 21 of 22 |
| C | 88 | 5 of 5 | 5 of 5 |
| D | 93 | 9 of 12 | 11 of 12 |
| E | 93 | 10 of 13 | 11 of 13 |
| F | 99 | 3 of 3 | 3 of 3 |
| All | 98 | 79 of 85 | 79 of 85 |
The thresholds follow Google's guidance: a good Cumulative Layout Shift is 0.1 or less, and Lighthouse rates mobile Total Blocking Time of 200 milliseconds or less as fast (web.dev, 2024; Chrome for Developers).
Field data told a different story from the lab. Of the 14 pages with real-user data, 12 passed Core Web Vitals and 2 failed. Both failures came from responsiveness (INP) and layout shift (CLS), not from loading speed. Half of the 12 passing pages scored below 75 in the lab, because real visitors often have faster phones and connections than the emulated test, and may have parts of the site cached.
Repeated runs showed how much a single test can mislead. The 13 sites tested three times moved between 0 and 31 points from run to run, with a median of 4. Two of them landed on both sides of 90 depending on the run. The lightest pages were the most stable: the three repeated sites with a median of 95 or more varied by 2 points at most.
What are the limits of this website speed study?
It is a small, non-random snapshot measured in a lab, so read the results as a clear pattern rather than a precise statistic for all SME websites.
- Small sample. 85 pages in total and only 3 to 30 per group. One site can move a group's median.
- Lab data. PageSpeed Insights simulates one device on one connection. Real visitors have different phones and networks, and only 14 pages had real-user data.
- Groups are not random. We chose the agencies from our own market research, and the client sites come from their portfolios, which agencies pick to show their best work.
- Mostly single runs. 72 of the 85 pages were tested once, and repeated runs varied by up to 31 points.
- Mostly home pages. We tested the page an agency or client links to, usually the home page. Other pages on the same site can score differently.
- One night. Sites change. A page tested on 28 October 2026 may score differently today.
- Agency type is not technology. A hand-coding studio's client site can still carry heavy images, and a builder site can be fast. The groups describe who built the site, not exactly how.
How should you cite this study?
Please cite it as follows and link to this page:
Visuco (2026). SME website speed study 2026: Google PageSpeed Insights results for 85 websites. Measured 28 October 2026, published 31 October 2026. /research/website-speed-study-2026/
You're welcome to quote the figures on this page, with a link to it. For questions about the method or the data, email sulo@aijagroup.fi.
Read what the numbers mean in practice in what is a good PageSpeed score? and see how we put our own sites to the same test in our speed guarantee.
Questions about the study
Why did you group the sites by agency rather than by technology?
Because a business owner chooses an agency, not a framework. Technology detection from page source is also less reliable than knowing who built the site.
Why don't you name the sites?
The study is about patterns, not about singling out other agencies or their clients. We publish only aggregated figures.
Why use the lab score instead of real-user data?
Only 14 of the 85 pages had enough real visitors for Google's field data. The lab score is the one measure available for every page, and anyone can repeat it.
Will you repeat the study?
We plan to measure again with more sites and three runs for every page, and to publish the results on this page with a new date.
Sources
- Google for Developers, "About PageSpeed Insights", accessed October 2026 developers.google.com/speed/docs/insights/v5/about
- web.dev, "Web Vitals", October 2024 web.dev/articles/vitals
- Chrome for Developers, "Total Blocking Time", accessed October 2026 developer.chrome.com/docs/lighthouse/performance/lighthouse-total-blocking-time