Blog · 5 November 2026 · 7 min read

Why your website is slow on mobile (and what to do about it)

“It's fast for me” is the most common reaction from an owner who sees a red PageSpeed score. And it is usually true: on a work laptop, on fibre, with the site already cached from last time. Your customer opens the same page on a three-year-old phone, on a bus, for the first time. This article explains where the gap comes from, what most often ruins the score on small-business sites, and where to start.

A phone is not a small laptop

Three things stack up. First, the processor: a typical mid-range Android phone runs a page's scripts four to eight times slower than a laptop. Every slider, visitor counter, chat widget and cookie plugin is code the phone has to chew through before it lets anyone tap anything. Second, the network: 4G in a city centre can be quick, but it has high latency and wild swings, and every extra file is another round trip. Third, the first visit: you have the site in your browser's cache; the customer downloads everything from scratch.

Google knows this, which is why it has judged sites by their mobile version since 2018 (mobile-first indexing), and why Core Web Vitals are driven mainly by measurements from real users' phones. The “Desktop” score in PageSpeed is nice to look at, but rankings and conversions follow the other one.

The five causes we see most often

This is not theory. It is what keeps coming up in our scans of service businesses, small shops and clinics, ordered by how often it appears.

They almost always come together. A typical page-builder business site has a 2 MB hero photo, eight plugins, three fonts and an embedded map, and takes 7–9 seconds to load on a phone while looking fine on the owner's laptop.

How to check it yourself in two minutes

1. Go to pagespeed.web.dev, paste your address and wait. Look at the “Mobile” tab, not “Desktop”.
2. If there is a “Discover what your real users are experiencing” section at the top, those are measurements from your customers' phones over the last 28 days, and they are what Google uses. If it is missing, the site has too little traffic and you are left with the lab measurement below.
3. Scroll to “Diagnostics” and look for the messages from the table above. Each one expands into a list of files, which is a ready-made to-do list for your webmaster.
4. Repeat the test for a service page or the contact page, not just the home page. That is often where it is worst.

Our free quick scan does the same in one step (a mobile Lighthouse run) and adds the security and SEO checks PageSpeed does not mention.

What order to fix things in

If the budget is limited, and it usually is, order matters. This is the sequence that gives the most for the least in practice:

1. Images. The cheapest fix with the biggest effect. Shrinking a hero photo from 3 MB to 150 kB alone can cut 2–4 seconds on 4G.
2. Caching and a CDN. An hour's work, often free (Cloudflare's free plan, for instance). It cuts server response time for every visitor.
3. A script clear-out. Go through the plugin and widget list and ask one question: “has anyone used this in the last quarter?” Half of it usually goes with no loss.
4. Fonts. A small thing, but it removes layout shifts and one or two files from the critical path.
5. Hosting. Only if TTFB is still above a second after the four steps above. Moving hosts without doing 1–4 rarely changes much.

What not to do

• Don't buy a “speed plugin” blind. Good caching plugins help, but they will not fix a 4 MB photo or twenty scripts. Diagnose first, then pick the tool.
• Don't chase 100. The difference between 92 and 100 is invisible to a customer; the difference between 35 and 75 is enormous. The goal is the green Core Web Vitals range, not a perfect score.
• Don't switch off things that sell. If the chat widget brings enquiries, keep it, but make it load after the first user interaction instead of in the first second.
• Don't commission “optimisation” without a list. The contractor should get specific items from the diagnostics, and you should get a measurement before and after. Otherwise you are paying for a promise.