How fast are business websites in Gdańsk, Gdynia and Sopot? I tested 100 sites in PageSpeed Insights.

Business owners often tell me their website “is fast”. Usually they check it on an office computer with a fast connection. I wanted to see what it really looks like, so I picked 100 business websites from Gdańsk, Gdynia and Sopot, the three cities that make up Poland’s Tri-City, and ran each one through Google PageSpeed Insights three times, separately for mobile and desktop. Below are all the results. On the charts you can switch devices, metrics and groups yourself.

Key findings

62median PageSpeed score on mobile

84median on desktop

6/99sites with a good mobile score (90-100)

7.7 smedian mobile LCP, against Google's 2.5 s threshold

On desktop most sites do reasonably well: 40 of 99 score in the good range and the median is 84. Mobile is a different story. 83 sites sit in the middle, orange band, 10 score poorly and only 6 make it into the green.

How I ran the study

• Sample: 100 business websites from 6 industries: architecture and interiors, HVAC, accounting, cleaning, dentistry and language schools. 54 sites from Gdańsk, 36 from Gdynia and 10 from Sopot.
• Selection: purposive, not random. The results describe these industries, but are not representative of every business in the Tri-City.
• Measurement: Google PageSpeed Insights API, Lighthouse 13.5.0, all tests on 26 October 2026. Each site was tested three times on mobile and three times on desktop, and I used the median.
• Completeness: 596 of 600 runs succeeded. 99 sites have complete series. One site showed a loading screen instead of content in its screenshots, but excluding it does not change the medians.
• Anonymity: sites are numbered S-001 to S-100. I do not publish names or addresses, because the point is a picture of the market, not calling out individual businesses.

These are lab test results, not a Core Web Vitals assessment for real users. I explain the difference below, because it matters.

Score distribution: mobile and desktop

On mobile the scores cluster between 50 and 69 points: 65 of 99 sites land there. That is a typical picture for a site built on a popular theme, with large images and a handful of plugins. Not dramatic, but a long way from a good score.

Every site on one chart

The gap between devices does not mean sites have a “worse mobile version”. The PageSpeed mobile test simulates a mid-range phone on a slow mobile connection. Every multi-megabyte image and every chat widget or ad pixel costs several times more time than on desktop. And most customers of local service businesses search on their phones.

Where sites lose the most time

The PageSpeed score is a weighted average of several measurements. To understand what drags it down, you need to look at the individual metrics:

• LCP (Largest Contentful Paint): when the largest element on screen appears, usually an image or heading. Good: up to 2.5 s.
• TBT (Total Blocking Time): how long the browser is busy with scripts and cannot respond to clicks. In the lab test it stands in for INP. Good: up to 200 ms.
• CLS (Cumulative Layout Shift): how much elements jump around while loading. Good: up to 0.1.
• FCP (First Contentful Paint): when anything at all appears on screen. Good: up to 1.8 s.

The key takeaway: on mobile the problem is almost entirely content loading. 84 of 99 sites have an LCP above 4 seconds, and 38 are above 10 seconds. Scripts (TBT) and layout shifts (CLS) do much better. Only 5 sites meet all three main thresholds on mobile, compared with 48 on desktop.

One curiosity: CLS is worse on desktop than on mobile. 16 sites exceed 0.25 on desktop, most often because of images without declared dimensions, sliders and late-loading fonts.

Industries and cities

Differences between industries on mobile are small: medians range from 60 to 66 points. Two things stand out. Accounting firms have by far the best desktop results (median 94.5, with 12 of 20 sites in the green), which fits their simpler, less image-heavy sites. Language schools have the slowest mobile LCP, with a median of 11.7 s.

Groups range from 10 to 54 sites, so this is a descriptive comparison. I would not conclude that any city “builds better websites”.

What slows sites down most often

The same problems repeat on almost every site, regardless of industry. Here is what is behind them and how to fix them:

Render-blocking resources (97% of sites)

CSS and JavaScript files loaded in the page head that the browser has to wait for before showing anything. Usually theme and plugin styles and libraries that the page does not even use. The fix: combine and slim down styles, load scripts with the defer attribute and inline critical styles.

Short cache lifetimes (92% of sites)

The server tells the browser to “forget” files quickly, so they are downloaded again on the next visit. Usually a single server rule or a Cloudflare setting. The cheapest fix on the list.

Late LCP resource discovery and image delivery (91% and 86% of sites)

The main image is too large, in an old format or lazy-loaded even though it is visible right away. WebP or AVIF, several image sizes for different screens, high fetch priority for the main image and no lazy loading at the top of the page all help. This is usually the biggest LCP win on mobile.

Unused JavaScript and CSS (88% and 81% of sites)

Code from plugins, sliders, page builders and widgets loaded on every page, even where it is not needed. The fix is to remove unnecessary plugins, load scripts only where they are used and, when rebuilding the site, choose a lighter technology.

Images without dimensions (69% of sites)

When an image has no width and height, the browser does not reserve space for it and pushes the rest of the content down once it loads. Hence jumpy pages and a higher CLS. The fix is simply adding width and height attributes.

Only one site in four has real-user data

PageSpeed Insights shows two kinds of data: the lab test, which is everything above, and data from real Chrome users from the CrUX report. The latter is what counts when Google assesses Core Web Vitals, but it only appears with enough traffic.

In this study 26 of 99 sites had page-level data in the mobile test and 9 in the desktop test. Origin-level data appeared for 24 sites on mobile and 9 on desktop. Most business websites in the Tri-City simply do not get enough Chrome traffic for Google to assess them based on real user behaviour. Missing data does not mean a bad result. It only means the lab test is the only guide you have.

Why it matters for SEO and AI visibility

• Google: Core Web Vitals are one of the signals used in ranking, but Google stresses that good content matters more. A fast site will not beat a better answer from a competitor. A slow site can, however, lose out where content is comparable, which is common in local services.
• Customers: someone looking for a dentist or a cleaning company on their phone compares several sites in a row. If the main content appears after 8 seconds, some people go back to the results before they see your offer.
• AI search: ChatGPT, Perplexity and AI Overviews rely on content that can be fetched and read quickly. Pages that only show content after heavy scripts load are harder to process, and many AI bots do not run JavaScript at all. A lightweight site with content in the HTML is the foundation of both technical SEO and GEO.

Compare your site

Where to start

If your site looks like most of the sites in this study, you do not need a new website straight away. The order that usually gives the most for the least effort:

1. Set cache lifetimes on the server or in Cloudflare.
2. Resize the main images and convert them to WebP, and give priority to the image at the top of the page.
3. Add dimensions to images.
4. Review plugins and scripts: remove unused ones and load the rest only where they are needed.
5. Only if that is not enough, consider rebuilding on a lighter technology.

After the changes, check not only PageSpeed but also the Core Web Vitals report in Google Search Console. If your site has real-user data, that is where you will see whether the improvement reached real visitors.