jonas.splinter
Search
Blog

Lighthouse on every page type

I ran Lighthouse on 13 page types, mobile and desktop, through the PageSpeed Insights API. Mostly 100, two 99s and one 66 on purpose, and why 100 is not the whole story.

This site got a new look today. After an accessibility sweep of the redesign, I also ran Lighthouse on every page type, on mobile and on desktop: 26 runs. Almost everything scored 100. The few scores below 100 have a reason I can explain, and I chose to keep them. This post covers the runs, the exceptions, a tip for the API and why a perfect score is not where the work ends.

26 runs through the PageSpeed Insights API

I ran Lighthouse through Google's PageSpeed Insights API (opens in a new tab), which runs it on Google's side. That means the results do not depend on my own machine or browser. I tested 13 page types, each on mobile and desktop, so 26 runs in total. The results:

  • Accessibility, best practices, SEO: 100 on every page, with one intended exception, explained below.
  • Performance: 100 everywhere, except 99 for the homepage and a game page on mobile.

A tip if you do the same: without your own key, the API uses a shared free quota that can run out, and then every request returns HTTP 429. A free key of your own, restricted to that API, fixes it.

Why the two 99s and the 66 are fine

The 99s come from Speed Index, at 2.6 and 3.0 seconds. Speed Index looks at how quickly the visible part of the page fills in, and on these two pages the tiles fade in one after another. The other numbers on those pages are clean: first content appears at 0.9 seconds, total blocking time is 0 ms and layout shift is 0.

So the page is quick to show something and does not jump around; the score drops one point because the animation takes its time. I kept the animation: it is part of the design. Removing it to gain one point would trade the look of the site for a number.

Single update pages score 66 on SEO because they are set to noindex on purpose: they are two sentences long. Lighthouse reads noindex as a page that cannot be indexed and marks it down, which is correct from its point of view. From mine, a two-sentence page is not something I want in search results on its own. The score reflects a decision, not a fault.

That is the general habit I try to keep with these reports: read the reason behind anything below 100 before changing anything. Sometimes the reason is a real problem, and sometimes it is a choice you made and are happy with.

A 100 is not the whole story

The bigger lesson: a 100 does not mean a site is accessible. Automated checks cover only part of WCAG. In the separate accessibility sweep of the same redesign, the most telling finding there came from a manual keyboard test on a phone: with the phone menu open, the page behind it could still be reached. No checker had found that. Contrast was also measured for every colour pair in the stylesheet, in both themes, not only for what Lighthouse happened to see on the screen it loaded.

So I read the Lighthouse scores as a floor. They confirm that nothing obvious is broken in performance, best practices or SEO, and they catch regressions cheaply. They do not replace a person with a keyboard.

What I do not know yet

I have not tested with a real screen reader. 200% zoom and text spacing were checked from the CSS, not in a browser by a person. Lighthouse does not answer either question for me.

Frequently asked questions

Why use the PageSpeed Insights API instead of Lighthouse in my own browser?

It runs Lighthouse on Google's side, so the results do not depend on my own machine or browser, and it is easy to repeat for every page type on mobile and desktop.

Why did every request return HTTP 429?

Without your own key the API uses a shared free quota that can run out. A free key of your own, restricted to that API, fixes it.

Does an accessibility score of 100 mean a site is accessible?

No. Automated checks cover only part of WCAG. On this site a manual keyboard test on a phone found a problem that no checker had found.

What to do first

  1. Run Lighthouse for mobile and desktop on each page type, not only on the homepage.
  2. Use a free key of your own for the PageSpeed Insights API, restricted to that API, so a shared quota cannot return HTTP 429.
  3. Read the reasons behind anything below 100, and keep a lower score when it reflects a deliberate choice.
  4. Treat 100 as a floor: tab through every page type without a mouse, including on a phone with the menu open.

The full accessibility work is in measuring accessibility after a redesign. How my projects now post updates here is in my projects post their own updates.

Related updates

Short notes on the same topics