Myths

Google Lighthouse accessibility score: what a 100 counts, and what to test by hand

6 min read
Mikhail ShaymardanovWeb Quality and Accessibility Expert

Right below the Google Lighthouse accessibility score, a store owner reading the report finds ten checks to do themselves, and the score counts none of them. Lighthouse, the engine behind PageSpeed Insights, grades the page on rules a program can judge alone: 58 at most in version 13.6. A 100 covers those rules, and everything else gets checked by hand, on the same page.

A store scored 100, zoomed to 200%


#How the Google Lighthouse accessibility score is calculated?

PageSpeed Insights, Google's online test, runs Lighthouse on the URL it's given, for mobile and desktop. Lighthouse takes its rules from axe-core, a set of automated tests, and each rule looks in the code for a flaw that gets in someone's way. One checks that the viewport tag, which sets how the page displays on a phone, still lets people with low vision zoom in.

The score is a weighted average. In Lighthouse 13.6, the version we tested with, 58 rules count toward it, each weighted 10, 7, 3 or 1 by severity. A single faulty element fails its rule for the whole page, with no partial credit. A rule with nothing to check drops out of the calculation, so a page without input fields is never graded on them.

Lighthouse judges the page right after it loads, on one simulated device: a 412 by 823 pixel phone or a 1,350 by 940 desktop window. Below the score, the report lists ten checks under “Additional items to manually check”, with no weight. Lighthouse describes them as areas an automated tool can't cover, like keyboard access to every control.

We ran Lighthouse on the home page of Havoc & Pawns, a demo board game store. On a first visit, a cookie consent dialog opens over it. On mobile and desktop, the home page scores 100, and only 27 rules find anything to judge. The other 31, plus five rules that carry no weight, land under “Not applicable”.

Screenshot: an English Lighthouse report for the demo store's home page, with a green gauge at 100 above the heading “Accessibility” and its description, then three collapsed groups: “Additional items to manually check (10)”, “Passed audits (27)” and “Not applicable (36)”.Marker 1. The accessibility score: 100. Marker 2. The “Additional items to manually check” group, ten checks, collapsed, with its sentence on what an automated tool cannot cover.
On the Havoc & Pawns home page, with the cookie dialog open, Lighthouse 13.6 shows 100marker 1 and sets ten manual checks apartmarker 2, which don't count toward the score.
Shopify theme accessibility: what Shopify checked, and what's left to test

#How to test a page that scores 100: zoom to 200% in Chrome?

The test takes no tool: in Chrome, hold Ctrl (⌘ on a Mac) and press the plus key until the zoom reaches 200%. Then use the page the way a person with low vision does when they zoom in to read.

WCAG 2.2 makes this its success criterion 1.4.4: text must resize up to 200% without losing content or functionality, captions and images of text excepted. Any one browser feature that gets there is enough, whole-page zoom included. EN 301 549, the European standard written to help businesses meet the European Accessibility Act, copies the criterion into its requirements for web pages.

#What happens to the demo store at 200%?

In our Chrome window, the page gets 1,100 by 633 pixels below the address bar. At 200%, that drops to 550 by 316, since each page pixel is now drawn twice as large. The cookie dialog, fixed in the middle of the screen, is then 587 pixels tall. Its title ends up above the top edge, and its two buttons fall below the bottom edge.

Screenshot: the Havoc & Pawns home page zoomed to 200% in Chrome; the cookie dialog fills the screen, starts partway through its paragraph with no title, and ends on the “Advertising” option, with no button in sight.Marker 1. The top of the screen: the end of the paragraph, with the title “Your cookie choices” above the edge. Marker 2. The bottom of the screen: the “Advertising” option, with the buttons below it.
Zoomed to 200% in Chrome, the cookie dialog runs past the screen: its title sits above the top edgemarker 1, and its “Reject all” and “Accept all” buttons sit below the “Advertising” optionmarker 2.

The mouse wheel scrolls the page behind the gray overlay, never the dialog. The Tab key does reach “Reject all” and “Accept all”, except that both stay off screen, so a keyboard user presses a button they can't see. With a mouse, a person who zoomed in can neither accept nor reject, and the store stays stuck behind the overlay.

We ran the test again, this time doubling only the text size in Chrome's font settings, and the buttons still left the screen. Whole page or text alone, the dialog fails the 200% test, because it's fixed in the middle of the screen and its content doesn't scroll. Zoom into the demo page yourself to see it.

#Why the score stays at 100?

In both screens Lighthouse simulates, the cookie dialog fits in full, with named buttons and enough contrast, so every rule passes.

The cookie dialog fits in both of Lighthouse's screens and runs 271 pixels past Chrome's screen zoomed to 200%.
ScreenVisible height, in page pixelsCookie dialog height
Phone simulated by Lighthouse823749
Desktop simulated by Lighthouse940564
Chrome window zoomed to 200%316587

The simulated phone offers 823 pixels of height, Chrome's zoomed screen 316: just over half the cookie dialog. A 100 on mobile therefore says nothing about zoom. Zooming in shrinks the height of the screen too, and height is what runs out here.

The only rule in the score that touches zoom checks that the viewport tag doesn't block it, and the store doesn't. None of the rules looks at the page once it's zoomed. We then made the dialog scroll whenever it runs past the screen (fixed version). At 200%, the mouse wheel and the Tab key now bring its buttons into view. Lighthouse still gives it 100: the score can't tell the two versions apart.


#What to do with a good accessibility score on your website?

File the dated report with what a person checked by hand on the tested page. The 100 covers what a program read in that one page's code when it loaded, as the bottom of the report itself shows.

Screenshot: the bottom of an English Lighthouse report, with the capture date and time, “Emulated Desktop with Lighthouse 13.6.0”, “Single page session”, “Initial page load”, “Custom throttling” and “Using HeadlessChromium 139.0.0.0 with cli”.Marker 1. The date and time of the test. Marker 2. The version: Lighthouse 13.6.0, with desktop emulation. Marker 3. The line “Single page session”.
The bottom of the report says what the 100 covers: the datemarker 1, the Lighthouse versionmarker 2 and a single page sessionmarker 3, to keep with the tests done by hand.

#What the report proves for a store the law covers?

Directive (EU) 2019/882, the European Accessibility Act, covers online stores that sell to consumers too, through each member state's own law. Since June 28, 2025, a covered store has had to explain how its service meets the accessibility requirements, in its general terms and conditions or an equivalent document. It also has to keep that information for as long as the service runs, and have procedures in place to stay compliant.

When a national authority that checks compliance asks, giving its reasons, the store has to hand over all the information needed to show its service complies. A microenterprise that provides services is exempt from the accessibility requirements and from these obligations. A microenterprise is a business that employs fewer than 10 people and has an annual turnover or balance sheet total of €2 million or less (who is covered).

A Lighthouse score and an audit's verdict count different things, and neither converts into the other. To talk about compliance, an audit checks every level A and AA criterion on a sample of pages, every page of the checkout included, as WCAG's evaluation method lays out. A page conforms at level AA only when it meets all of them, with no weights and no average.

Website accessibility audit: cost, quotes and how to choose an auditor

#What to ask the agency next?

Start by asking the web agency to fix whatever the score flags. Then have it run the 200% zoom test and the keyboard test on the home page (cookie dialog included), a product page, the cart and checkout. For each page, it notes the date, the tool, its version and what was checked by hand.

For Havoc & Pawns, our demo store, the file for the home page before the fix fits on two lines: the Lighthouse 13.6.0 report, 100; the cookie dialog at 200%, buttons off screen, to fix. The first line comes from a program, the second from a person, on the same page.

Also ask what was checked by hand among the ten “Additional items to manually check” in the report, and what is left to do. To talk about compliance, add a quote for an audit on a sample of pages, checkout flow included.


The zoom test needs no tool. On a live page, Includdy's scan lists the accessibility issues it detects. Its guided manual tests go through headings, images, lists, links and buttons, colors and required elements, with a suggested verdict for each item that a person confirms or corrects. For compliance, Includdy also carries out accessibility audits.


Share this article

Frequently asked questions

PageSpeed Insights rates a score of 90 or above as good, 50 to 89 as needing improvement, and below 50 as poor. A failed rule scores zero and still counts at its full weight in the average, whether one element on the page fails it or a hundred.

PageSpeed Insights runs Lighthouse, whose report lists ten “Additional items to manually check” below the score, with no weight: Lighthouse leaves them to a person.

Only in part. Lighthouse takes its rules from axe-core, and a rule can carry the label of a WCAG success criterion. The viewport rule, which checks that the page can be zoomed, is tagged with criterion 1.4.4, Resize Text.

A WCAG audit judges every level A and AA success criterion on a sample of pages, every page of the checkout included. Lighthouse judges automated rules on the one URL it is given.

It depends on what is counted: the figures people quote don't measure the same thing. The UK government's accessibility team planted 142 barriers in one page, and in 2018 the best of the 13 tools it tested found 40% of them.

In a 2021 report, Deque, the company behind axe-core, set out to disprove the widely held belief that automated testing covers only 20 to 30%. It counted 57.38% of the issues in its own audits as found by its automated tests.