Research

Measurements from our own audit base — every site we run through WebSpeed leaves behind its markup, its filter log and its before/after metrics. These are the questions we could answer with that, and nobody else could.

9 sites · measured 26 Aug 2026

We cut the image bytes in half and the score did not notice

We re-encoded every third-party image on nine sites to WebP and served it ourselves. Median saving 43% per file. The mobile PageSpeed score moved by a median of zero, and not one site moved further than its own run-to-run noise.

1 of 21 images that arrived sooner through our own endpoint than from the original CDN
428 sites · measured 24 Aug 2026

Five sites in 428 serve HTML uncompressed. Which algorithm the rest use was decided by their platform

We re-requested the home page of every site in our audit base with a browser's Accept-Encoding header. 97% return compressed HTML, and only five real pages come back raw. What varies is the algorithm, and on hosted platforms it is unanimous per platform, while WordPress splits three ways.

13 of 428 sites returned HTML with no compression at all
440 sites · measured 24 Aug 2026

93% of reports say to remove unused JavaScript. On the same 440 sites, what we could change was the order of delivery

We counted every hint PageSpeed returned for 440 audited sites and put it next to what our proxy was able to change on the same sites. The two lists disagree at the top — the most common advice is the advice that requires editing the site's own bundles, and a third of the report describes one cost under three names.

93% of reports flag unused JavaScript
447 sites · measured 24 Aug 2026

Measured three times in one sitting, 18% of mobile scores land in two different bands

We measure every page three times per run and store each round. Across 447 audited sites, half the rounds that actually executed disagreed by 8 points or more, 18% of mobile measurements crossed a Lighthouse band inside one sitting, and a quarter of before/after comparisons changed sign between two measurements days apart.

18% of mobile measurements put two rounds in different Lighthouse bands
445 sites · measured 19 Aug 2026

Zero of 445 audits report render-blocking resources. Of the ones we re-read, 78% still block the render

In Lighthouse 13 the render-blocking-resources audit is gone from PageSpeed responses, replaced by render-blocking-insight. Across 445 audited sites, our stored data showed the problem at zero — until we changed how we read the response and found it on 78% of them.

0 of 445 audited sites carry a classic render-blocking-resources key
452 sites · measured 12 Aug 2026

More font files, lower score — except on the platforms that choose the fonts for you

We counted the font files on 452 audited sites. The median loads five, they are 6% of the page weight, and the count still sorts the sample by PageSpeed score. On the hosted builders, where the platform picks the fonts, that pattern falls apart.

5 font files the median site loads
482 sites · measured 4 Aug 2026

63 sites carry Cloudflare's opt-out attribute. Two of them run Rocket Loader

data-cfasync="false" tells Cloudflare Rocket Loader to leave a script alone, and plenty of tools read its presence as evidence Rocket Loader is running. On 63 audited sites carrying the attribute, a live browser-agent request found the Rocket Loader runtime on two.

63 of 482 audited sites ship the data-cfasync attribute
457 sites · measured 4 Aug 2026

On 45% of the sites we audited, JavaScript outweighs the images

We took the full resource inventory of 457 audited sites and added up what each page is made of. The median page is 3.4 MB over 102 requests, and the single heaviest resource type is more often JavaScript than images.

3.4 MB median total page weight, over 102 requests
457 sites · measured 4 Aug 2026

The median site we audited loads from 12 hosts it doesn't own

We counted the distinct hosts each of 457 audited sites connects to before it finishes loading. The median is 12, a quarter of sites pass 20, and the count sorts the sample by PageSpeed score almost perfectly.

12 distinct hosts the median site loads from, beyond the one serving the page
381 sites · measured 4 Aug 2026

Above 70, speeding up a site is a coin flip

We compared the mobile PageSpeed score of 381 sites before and after routing them through our optimizing proxy. It rose on 67%. But the starting score decides almost everything — under 50 it nearly always helps, over 70 it is even odds.

67% of sites scored higher on mobile through the proxy
261 sites · measured 26 Jul 2026

On 37% of sites, the browser can't find the hero image in the HTML

Every LCP guide says to preload your hero image. We checked 261 audited sites for whether a hero image is discoverable in the HTML the server sends at all. On 96 of them there was nothing to preload — and the reason splits the web in two.

63% of sites had a hero image the browser could discover in the markup
9 sites · measured 23 Jul 2026

A caching plugin didn't make these WordPress sites fast on mobile

We measured nine WordPress sites run by the industry itself — plugin makers and publishers who write about speed. The median mobile PageSpeed was 51. Where we could confirm a caching plugin was installed, the site was still slow. Here's the part caching can't reach.

51 median mobile PageSpeed across the nine sites, before
168 sites · measured 21 Jul 2026

Online stores don't get slow differently from blogs — we checked

We sorted 168 audited sites into stores, publishers, landing pages and corporate sites to see whether each type gets slow its own way. Every difference disappeared once we controlled for the platform.

4 of 4 apparent differences vanished when platform was held fixed
162 sites · measured 21 Jul 2026

Half the sites we audited ship images without dimensions — and their CLS score says they're fine

We parsed 162 audited sites. 53% ship at least one image with no width and height. 95% of those still get a "good" Cumulative Layout Shift score in the lab.

53% of sites ship at least one image with no dimensions

One rule for this section: a page only belongs here if the finding comes out of measurements we ran ourselves, on a stated sample, on a stated date. Every study prints its method and what it doesn't show. Null results get published too — the biggest sample here, 482 sites, produced one.

Your site, measured the same way

A free before / after PageSpeed comparison on mobile and desktop. No signup.

No signup. See your before/after PageSpeed score in about a minute.