Website speed comparison: we measured 40 agency websites.

Every agency in this study sells search performance. Most of them sell page speed explicitly — it is on their services pages, in their pitch decks and in their own blog posts about Core Web Vitals. So we measured their websites in a straight website speed comparison. Not their clients’ websites: their own. 37 of the best-known SEO and digital agencies in Germany and Romania, plus our own three domains, 40 sites in total, each measured twice over: what the homepage weighs, and what Google’s own PageSpeed Insights says about it across three consecutive runs.

Google PageSpeed Insights speed score, mobile

Every bar is one homepage on Google’s PageSpeed Insights, scored out of 100 on its mobile test. Google paints anything below 50 red and anything below 90 amber: only 3 of the 37 agency sites reach green. Select a bar for what that site weighs. The three studies at the end are other people’s work; their figures are quoted as printed on the pages linked.

What we measured, and how many times

We loaded 40 homepages: 22 German agencies, 15 Romanian ones, and our own three domains. Each was measured twice over. Lighthouse 12.8.2 on its mobile preset gave us total page weight and the number of network requests. Google’s PageSpeed Insights gave us the four scores it publishes — performance, accessibility, best practices and SEO — and we ran it three times per site.

Page weight and request count are properties of the website. They are what the server chose to send. Run the test from Bucharest or Berlin, on a busy laptop or an idle one, and you get the same answer. That is why they lead the table.

A PageSpeed performance score is not like that, and this is the part most comparisons get wrong. It is lab data measured on Google’s machines, under whatever load those machines are under when your request arrives. Webdigital returned 77 / 63 / 46 across three consecutive runs — a 31-point spread on a page that did not change between them. 18 of the 40 sites moved ten points or more between their best and worst run.

So every PageSpeed number in this article is the median of three runs, and the table says so. If you run one of these sites yourself and get a different number, that is the tool, not a contradiction — run it three times and compare the middle value. The other three categories behave differently: accessibility, best practices and SEO are deterministic rule checks, and they returned the same score on nearly every pass.

Everything here was measured on 7 September 2026. Sites change. If you are reading this months later, re-run it — the method is in the sources at the bottom and takes about two minutes per site.

Page weight: the lightest site weighs 322 KiB, the heaviest 189.5 MB

The median agency site in this study ships 1,642 KiB across 92 requests. The German median is 1,522 KiB; the Romanian median is 2,376 KiB.

Wingmen at 342 KiB is the lightest of the rest, and it lands within 20 KiB of ours — proof that a light agency site is a choice rather than a constraint. At the other end, eprofessional ships 33.6 MB, and Kubis Interactive ships 189.5 MB across 879 requests — dozens of individual PNG files over two megabytes each, on a homepage. That is not a configuration mistake anyone made once; it is a site nobody has opened a network panel on.

For context, Kubis Interactive’s homepage alone would exhaust a 150 MB mobile data allowance in a single visit, with change owed. On a metered connection, opening it costs the visitor money.

What Google’s own speed test says about agency websites

The median competitor in this study scores 56 out of 100 for mobile performance. Google colours anything below 50 red and anything below 90 amber. 13 of the 37 agency sites — 35 per cent of them — sit in the red band, 21 sit in the amber, and 3 reach green at 90 or above. The best of them is Wingmen at 93; the lowest are Xivic and TUYA Digital, tied at 29.

The other three categories are the more damning ones, precisely because they do not move between runs. Accessibility, best practices and SEO are rule checks: the same page gets the same score every time, so there is no server load to blame. The competitor medians are 89 for accessibility, 96 for best practices and 100 for SEO.

3 of the 37 agency sites scored 100 on all three at once. SEO is where the field does best — the median there is 100, and that category is a floor rather than a ceiling: it checks that a page has a title and a meta description, that its links are crawlable, that its images carry alt text, and that canonical and hreflang are valid. Accessibility is where the field loses it, at a median of 89. That is not a ranking argument — it is the part of the audit that says some visitors cannot use the page at all.

Where we came out, and why that is not the point

Our three domains take the first three places on page weight: 322 KiB and 26 requests for the lightest, against a competitor median of 1,642 KiB and 92 requests. That is 5.1× lighter and 3.5× fewer requests than the middle of the field.

On PageSpeed our three domains take the first three places as well: levelui.com, levelui.ro and levelui.de each score 100 for speed, accessibility, best practices and SEO. Second place goes to Wingmen at 93, and the median agency site scores 56.

We are in the table because a study that measures everyone except its author is worth nothing. But the ranking is not the finding. The finding is that the median agency site is 5.1 times heavier than it needs to be, that the floor is already occupied by Wingmen, and that the categories nobody can blame on server load are the ones most of this industry is quietly failing. We later ran the same three-run method on the websites of Europe’s 50 biggest companies, whose median came out at 54.

There is nothing clever behind our own number. It is a full marketing site: a portfolio, a priced catalogue you can browse, a cart, three languages. It is light because every image is a modern format at a sane resolution, because the JavaScript that is not needed for the first screen is not downloaded for the first screen, and because nothing third-party loads unless a visitor asks for it. We wrote up the specifics in how to get a 100 PageSpeed score and what the metrics actually mean in Core Web Vitals explained.

What slow actually costs

Everything above is a measurement, and a measurement is not an argument. Weight matters because the money leaks in three separate places — and the one this industry names most often is the smallest of them.

The largest cost is the visitor who converts less, and it has been measured directly. Deloitte and Google instrumented 37 European brands and found that a 0.1-second improvement in mobile load time raised retail conversion rates by 8.4% and average order value by 9.2%; for travel, conversions rose 10.1%. A tenth of a second. The striking part is not the percentage but how small the intervention was that produced it.

Second is the visit that never turns into anything. Vodafone A/B tested two versions of the same page, one optimised for Core Web Vitals, and the version with a 31% better LCP produced 8% more sales and a 15% better lead-to-visit rate. That is a controlled test on one company’s own traffic rather than an industry average, which makes it the more useful of the two figures.

Rankings come third, and they are the smallest of the three. Core Web Vitals are a real ranking signal, and Google has been consistent that they are one signal among many: a fast page with nothing to say will not outrank a slow page that answers the question. Speed is a tie-breaker in the results and a gatekeeper on the visit. Most of this industry argues it the other way round.

For an agency site there is a fourth cost no study will measure. The homepage is the portfolio piece. A prospective client who opens a web agency’s own site on a phone, waits, and closes it has learned something about that agency no case study will unlearn — and a 33.6 MB homepage teaches it in the first few seconds.

One number is deliberately missing here. The most-quoted statistic in this field — that 53% of mobile visits are abandoned after three seconds — we could not trace to a primary source that still publishes it, so we left it out. The standard the rest of this article is held to applies to its arguments too.

How to check your own site in two minutes

Open your homepage in Chrome, press F12, go to the Network tab, tick “Disable cache”, and reload. The status bar at the bottom gives you both of the numbers this study leads with: how many requests, and how many bytes transferred. Compare that against the table above.

Then run [PageSpeed Insights](https://pagespeed.web.dev/) on the same URL three times and take the middle score. One run tells you very little — 18 of the 40 sites here moved ten points or more between their best and worst run.

If you are over a megabyte, the cause is almost always one of three things, in this order: uncompressed images, a font family loaded in six weights when the design uses two, and a tag manager pulling in analytics, chat, heatmaps and a cookie banner before anything is visible. In this study the heaviest sites were heavy for the first reason by a wide margin.

None of that needs a rebuild. It needs somebody to look. If you would rather we looked, our technical audit covers exactly this ground, and the SEO audit is free.

Method, tools and sources

Everything in this article was measured on 7 September 2026 and can be reproduced with freely available tools. The full list of sites is in the table above.

• Google Lighthouse 12.8.2 ↗

  Run from the command line against each homepage, mobile preset, one run per site. Page weight is the total-byte-weight audit; request count is the length of the network-requests list.

• PageSpeed Insights ↗

  Run three times per homepage on the mobile strategy, collecting all four categories. The published figure is the median of the three runs.

• Lighthouse performance scoring ↗

  Google’s own documentation on why performance scores vary between runs of the same page, and why a single run should not be treated as a measurement.

• Milliseconds Make Millions — Deloitte and Google ↗

  The 0.1-second study across 37 European retail and travel brands. The conversion and average-order-value figures quoted above are read from this page.

• Vodafone: a 31% improvement in LCP increased sales by 8% ↗

  An A/B test on Vodafone’s own traffic, measuring LCP on real user sessions with the PerformanceObserver API rather than in a lab.

• Understanding Google Page Experience ↗

  Google Search Central on where page experience sits among ranking signals — the basis for treating speed as a tie-breaker rather than a lever.