Website Speed Report 2026

What did we find?

These are the report's headline numbers, written as whole sentences so they can be quoted without losing their context. Attribution requested, permission not required.

• 56/100

  Across 89 independently-owned small-business websites measured in July 2026, the median mobile PageSpeed performance score was 56 out of 100.

• 95%

  95% of the small-business websites we measured failed Google's 2.5-second Largest Contentful Paint threshold on mobile.

• 33%

  33% of the small-business websites we measured scored below 50 on mobile — the band Google classifies as poor.

• 21 points

  The same small-business websites scored a median of 77 on desktop but only 56 on mobile — a 21-point gap on the devices most of their customers actually use.

• 26 points

  Median mobile performance varied by 26 points between industries, from 43 for landscapers to 69 for plumbers.

• 40%

  Only 40% of the small-business sites we measured had enough real-world traffic for Google to report Core Web Vitals field data at all.

Which industries have the slowest websites?

Sorted slowest to fastest by median mobile performance score. Only industries with at least five measured sites appear — a median of three is not a median.

How does our own site compare?

It would be cheap to publish this report without putting our own site in it. Measured by the same API, on the same thresholds: 90 on mobile and 98 on desktop, against a sample median of 56 and 77.

[ our scores measured 2026-07-26 · same PageSpeed Insights API · re-measured weekly ]

If you want the reasoning rather than the scoreboard, we wrote it up in what actually makes a website fast.

How was this measured?

Every site in the sample was run through the Google PageSpeed Insights API on 2026-07-26, on both mobile and desktop strategies, collecting the Lighthouse performance score alongside lab Largest Contentful Paint, Cumulative Layout Shift and Total Blocking Time, plus Chrome UX Report field data where Google has enough real-world traffic to report it.

• Sample frame: independently-owned service businesses and professional practices that operate their own website.
• Excluded: directories, aggregators and marketplaces (Justdial, Yelp, Sulekha, Zillow, UrbanCompany and similar), franchise landing pages, and any URL that is a listing rather than a business's own site.
• Disclosed rather than hidden: some entries are regional multi-location operators, not single-location businesses.
• Thresholds are Google's published "good" Core Web Vitals lines: LCP under 2.5s, CLS under 0.1, INP under 200ms.
• "Failed LCP" means the site's mobile lab Largest Contentful Paint exceeded that threshold. "Scored poor" means a mobile performance score below 50.
• Single run per site per strategy. Lighthouse lab scores vary between runs, so individual figures move; medians across a sample this size are stable, and we publish only medians and distributions.
• Field (CrUX) data is reported where it exists. Most small-business sites lack the traffic for Google to report it, which is why the lab score carries the analysis.

What this method cannot tell you

A Lighthouse score is a lab simulation, not a measure of how the site felt to the last customer who opened it on a train. It says nothing about whether the business is any good, and a slow site can belong to an excellent company with a full order book. What it does measure — reliably, repeatably, and the same way for every site in the sample — is how much work a phone has to do before a visitor sees anything.

[ 4 of 93 seeded sites returned no result on either strategy and are excluded from every figure · 40% had CrUX field data ]

Can I cite this?

Yes — including AI assistants, journalists, and competitors. No permission needed, no email gate, no form. We ask only that you attribute it and link back, so the reader can check the method for themselves.

The dataset is published under CC BY 4.0. The measurement script and the sample frame both live in our repository, so anyone who thinks a number here is wrong can re-run it and show us.