Testing accessibility: what tools find, and what only people find

TL;DR

Difficulty: Beginner

  • Automated checkers such as axe, Lighthouse and WAVE are free and fast, and they find the mechanical failures: contrast, missing labels, missing text alternatives.
  • They cannot judge everything. In a GDS test, 29% of known barriers were missed by every tool, and a Lighthouse score of 100 says nothing about keyboard use or focus.
  • A 20 minute manual pass (keyboard, zoom to 200%, reflow at 320 pixels, contrast by hand, forms) and a short screen reader pass find most of the rest.
  • Sessions with people who use assistive technology find what no checklist can.
  • Put axe checks in your automated tests, and repeat the manual pass on every release that touches the critical path.

You want to know whether a customer who is blind, or one who cannot use a mouse, can actually buy from your site. Here is how to find out without guessing: what the free tools catch, what a person catches in 20 minutes with a keyboard and a screen reader, and what only real users catch. One invented Galway bookshop, tested in one hour.

How do I know whether my site works for everyone?

You have built a shop, or paid someone to build it, and now a question sits in the back of your mind. Can a customer who is blind actually buy a book from it? Can someone who cannot hold a mouse get through the checkout? You do not want to guess, and you do not want a badge that says "accessible" on a site that is not.

The honest answer comes in three layers, and each one finds things the others miss:

  • Automated tools read the code of a page and flag the problems a machine can prove: missing labels, low contrast, images without text alternatives.
  • A manual pass is a person working through the site with the keyboard, the zoom and a contrast checker, looking for the problems only a human can judge.
  • People who use assistive technology every day show you whether the whole thing is usable, which no checklist can.

The W3C, the body that publishes the Web Content Accessibility Guidelines (WCAG), puts it plainly: no tool alone can determine whether a site meets accessibility standards, and knowledgeable human evaluation is required.1 This article shows all three in practice, and how to keep testing once the site is live. For the standard itself, read WCAG 2.1 AA explained. If someone has offered you a plugin that promises to do all of this, read why accessibility overlays do not make your site compliant first.

Example: the shop in this article

Claddagh Books is an independent bookshop in Galway with an online shop at claddaghbooks.example. Sinéad owns it. Ade is the developer who looks after the site. Sinéad has heard that accessibility is a legal duty and asks Ade a simple question: "Is our site accessible, and how would we know?" Ade blocks out one hour to find out. Each example that follows is one part of that hour.

Automated checkers read the code, not the experience

An automated accessibility checker is software that loads a page, reads its HTML, CSS and the result of its JavaScript, and tests it against a set of rules. Each rule is something a machine can decide with certainty. Does this image have an alt attribute? Does this form field have a label tied to it? Is the contrast between this text and its background at least 4.5 to 1? The answer is yes or no, so software can give it.

Three free tools cover most needs. Here is what each says about itself.

axe is an open source rules engine by Deque Systems. It runs in a browser extension and inside automated tests. Its own documentation says it finds on average 57% of WCAG issues automatically, and that it marks elements as "incomplete" where it cannot be certain and a person needs to look.2 That second part matters: a result of "needs review" is not a pass.

Lighthouse is built into Google Chrome's developer tools. Its accessibility score is a weighted average of pass or fail audits, with the weights based on axe's assessment of user impact. The same documentation lists a set of manual checks, such as a logical tab order and focus moving to new content, and says these do not affect the score.3 So a score of 100 means every automated audit passed, nothing more.

WAVE is a tool from WebAIM, a non profit at Utah State University. It draws icons straight onto your page to show errors and alerts in place. Its help page is refreshingly direct: "WAVE cannot tell you if your web content is accessible. Only a human can determine true accessibility."4

All three are fast, free and reliable on the mechanical failures. The W3C sums up their limit: tools cannot check all accessibility aspects automatically, sometimes produce false or misleading results, and can only assist a person in judging accessibility.5

Example: minutes 0 to 15, the automated scan

Ade picks the five pages a customer needs to buy a book: the home page, search results, a product page, the basket, and the checkout up to the payment step. He runs the axe browser extension on each, then Lighthouse, then WAVE, and writes every finding into one list, removing duplicates where two tools report the same thing.

After 15 minutes the list holds eight distinct issues:

  1. Grey price text on white, measured at 3.1 to 1, below the 4.5 to 1 that WCAG success criterion 1.4.3 asks for.6
  2. White text on the pink "sale" badge, at 2.9 to 1, the same criterion.
  3. The search box has a placeholder ("Search books") and no label, so a screen reader may announce only "edit text".
  4. The basket icon and the three social icons in the footer are links with no accessible name.
  5. The "sort by" dropdown on search results has no label.
  6. Author photos on product pages have no alt attribute at all.
  7. The checkout template is missing the page language, lang="en-IE", which success criterion 3.1.1 asks for.6
  8. On the product page the headings jump from H1 straight to H4, which WAVE flags as an alert to review.

Lighthouse gives the home page 84. Ade notes one more thing that none of the tools complained about: every book cover in the search results has the text alternative "book cover". It exists, so every tool passed it. Whether it helps anyone is a question for later in the hour.

How much the tools catch: two honest numbers

Two studies are worth knowing, because they measure different things and both are open about it.

In 2017 the Government Digital Service in the UK built a deliberately broken web page with 143 known barriers and ran ten automated tools against it.7 The best tool, counting errors and warnings, found 37% of the barriers. Counting its prompts for manual inspection too, the best found 41%. All ten tools together found 71%, and 29% of the barriers were missed by every tool. Their verdict: use the tools, never rely on them alone. Tools have improved since 2017, so read the numbers as history and the shape as current.

In 2021 Deque, the company behind axe, published a study of real first time audits: more than 13,000 pages and page states and nearly 300,000 issues.8 It found that 57.38% of the issues by count were found by its automated tests.9 The same report found automated issues for only 16 of the 50 success criteria at WCAG 2.1 Level AA.9 Deque is a tool vendor with an interest in a high number, and it says openly that it set out to challenge the usual estimate of 20% to 30%. Its data is still useful, because it shows why the two ways of counting differ. Both studies compute the same share:

c=ntn×100%c = \frac{n_t}{n} \times 100\%

where cc is the coverage as a percentage, ntn_t is the number of issues the automated tools found, and nn is the number of issues found by every method together. If percentages feel rusty, Foundations: percentages walks through them from scratch.

The difference lies in what you count as one issue. Low contrast text is one success criterion, but on a real shop it can appear 40 times, and a tool finds every one of them. Deque's table shows contrast alone made up 30% of all issues in its sample, and 83% of those were found automatically.9 Count every occurrence and the tools look strong. Count the kinds of problem and they look weak: in Deque's table, every focus order issue and every focus visible issue came from manual testing.9

Example: counting Ade's list both ways

By the end of his hour (the full list follows in the section on the result of the hour) Ade has 18 distinct issues, and the tools found 8 of them:

c=818×100=44.4c = \frac{8}{18} \times 100 = 44.4

So c≈44%c \approx 44\%: the tools found fewer than half of the distinct problems. Counted by occurrences, the picture flips. The grey price text alone appears on 46 product tiles, so the tools look as if they found most of the problems. Both numbers are true. The one that matters to Sinéad is the first, because each of the ten issues the tools missed can stop a customer on its own.

What the tools catch well, and what that tells you about the web

The mechanical failures are not rare. Every year WebAIM runs WAVE across the home pages of the top one million websites. In the 2026 report, 95.9% of those home pages had detectable WCAG failures, an average of 56.1 errors per page.10 Six kinds of error made up 96% of everything detected, and they have been the same six for seven years.10

Bar chart from the WebAIM Million 2026 report. Share of home pages with each error: low contrast text 83.9%, missing alternative text for images 53.1%, missing form input labels 51%, empty links 46.3%, empty buttons 30.6%, missing document language 13.5%. Together these six types account for 96% of all detected errors. Tap or click to open full screen and move around

That is the good news hidden in the bad. These six are exactly what a free tool finds in minutes, and fixing them is ordinary work. So run the tools first, fix what they find, and then spend your human time where the tools are blind.

Example: Ade's eight findings against the six

Seven of Ade's eight findings sit in the top six: contrast, missing alternatives, missing labels, empty links and the missing page language. Claddagh Books is completely normal. He fixes all eight the next afternoon: darker grey and pink, real <label> elements, names for the icon links, alt text for the author photos, lang="en-IE" on the checkout layout, and a corrected heading order. Lighthouse now shows 100.

A perfect score is the start of testing, not the end

A score of 100 feels like a finish line. It is the moment the real testing can begin, because the noise is gone and what remains is the work only a person can do. Lighthouse itself lists the checks it leaves to you: whether focus is trapped, whether interactive controls can be reached by keyboard, whether the tab order is logical, whether focus moves to new content added to the page.3 None of those affect the score.

Example: what 100 hides at Claddagh Books

After the fixes, the home page scores 100. Ade presses Tab once. Nothing visible happens. He presses it again and again and cannot see where he is, because a stylesheet sets outline: none on links to keep the design clean. A sighted customer who navigates by keyboard is now lost on a page that scores perfectly. The score was not lying. It simply never asked the question.

The manual pass: five checks a person can do in 20 minutes

The manual pass needs no special software. It needs a keyboard, the browser zoom, a contrast checker, and a little patience. Here are the five checks, each with the WCAG success criterion it tests.

  • Keyboard only. Put the mouse away. Press Tab to move forward, Shift and Tab to move back, Enter to follow links and press buttons, Space for checkboxes and buttons, the arrow keys inside menus and radio groups, Escape to close dialogs. Every function must work this way (2.1.1), focus must never get stuck (2.1.2), the order must make sense (2.4.3), and you must always see where focus is (2.4.7).611 My chapter on keyboard and focus goes deep on each.
  • Zoom to 200%. Text must resize to 200% without losing content or function (1.4.4).12 Use the browser zoom, not a pinch.
  • Reflow at 320 CSS pixels. Narrow the window, or set the browser's responsive mode to 320 pixels wide. Content must fit in one column with no sideways scrolling, apart from things like maps and data tables that genuinely need two dimensions (1.4.10). The W3C explains that 320 CSS pixels is what a 1280 pixel wide screen becomes at 400% zoom.13
  • Contrast where tools cannot measure it. Text over a photograph, a focus outline, the border of an input field, text in a hover state. Text needs 4.5 to 1, large text 3 to 1 (1.4.3), and the parts of controls you need to see need 3 to 1 against what is next to them (1.4.11).146
  • Structure and forms. Do the headings describe the sections (2.4.6)? Is every field labelled (3.3.2)? Submit a form with a mistake: is the error named in text and does it say how to fix it (3.3.1, 3.3.3)?6 My chapter on accessible forms shows the fixes.

The UK government's GOV.UK Service Manual gives the same advice to every public service team there: test regularly with both automated and manual testing, because you will miss some issues if you only do automated testing.15

Example: minutes 15 to 35, Ade's manual pass

Ade starts on the home page with the mouse pushed aside.

Keyboard. Focus is invisible on every header link (the outline: none from earlier), a failure of 2.4.7. On the product page he opens "quick view", and focus stays on the page behind the dialog, so the next Tab moves through links he cannot see, a failure of 2.4.3. In the checkout, the three delivery options (standard, next day, collect from the shop) are styled cards built from <div> elements that react to a click and nothing else. Tab skips straight past them. That is a failure of 2.1.1, and it blocks the sale completely.

Zoom and reflow. At 200% the sticky header, which holds the logo, the search box and the basket, takes up nearly half the screen. When Ade tabs into the checkout's Eircode field, the field sits underneath the header and he cannot see what he types. WCAG 2.2 names this exact problem in success criterion 2.4.11, Focus Not Obscured (Minimum), at Level AA.16 At 320 pixels the search results keep the filter sidebar next to the grid and the page scrolls sideways, a failure of 1.4.10.

Contrast by hand. The home page banner puts white text on a photo of the shop window. On the dark part of the photo it reads grand. Over the bright window it drops to about 2 to 1, which axe could only mark as "needs review" because it cannot see the pixels behind the text.

That is six new issues in 20 minutes, and every one of them passed the automated scan.

Screen readers: learn four commands, not forty

A screen reader is software that turns what is on the screen into speech or braille, and lets a person move through a page by its structure: headings, links, form fields, landmarks. Testing with one does not make you a screen reader user, and you should not pretend it does. Its job in your test is narrower. It tells you whether your code exposes the right names, roles and states, and whether the page makes sense read in order, without the layout.

Which ones should you test with? WebAIM's tenth screen reader user survey, run in December 2023 and January 2024 with 1,539 valid responses, gives the clearest picture.17 On desktop and laptop, 40.5% named JAWS as their primary screen reader, 37.7% NVDA and 9.7% VoiceOver. In Europe NVDA led JAWS, 37.2% to 29.7%. On mobile, 91.3% of respondents used a screen reader, and 70.6% named an Apple device as their primary mobile platform, against 27.6% for Android. When looking for information on a long page, 71.6% navigate by headings first.

The UK government's GOV.UK Service Manual asks its services to work with JAWS, NVDA, VoiceOver on iOS, TalkBack, a screen magnifier and Dragon speech recognition before public beta.18 For a small business, NVDA on Windows and VoiceOver on an iPhone cover a large share of real users at no extra cost if you own the devices.

Here are the few commands you need for a first test, taken from each vendor's own documentation.

NVDA on Windows (free, from NV Access). If you let the installer create the shortcut, Control+Alt+N starts it. The NVDA key is Insert by default. In a web page, H jumps to the next heading, 1 to 6 to the next heading at that level, F to the next form field, K to the next link and D to the next landmark. NVDA+F7 opens a list of all links, headings and landmarks. Control stops speech at once.19

VoiceOver on a Mac. Command+F5 turns it on and off. The VoiceOver modifier, written VO, is Control and Option together, or Caps Lock.20 VO+U opens the rotor, a menu of headings, links and form controls on the page.21 Control pauses speech.20

VoiceOver on an iPhone. Ask Siri to "Turn on VoiceOver", or use the Accessibility Shortcut if you set it up.22 Swipe right to move to the next item, swipe left for the previous one, double tap to activate the selected item, and swipe up with two fingers to read the whole screen from the top.23 Practise in Settings first: with VoiceOver on, the phone works differently.

TalkBack on Android. Hold both volume buttons for a few seconds to turn it on.24 Swipe right for the next item, double tap to activate it.25

Example: minutes 35 to 55, listening to the shop

Ade turns on NVDA, opens the search results for "Galway" and presses K to move through the links. He hears "book cover, link", then "book cover, link" again, twenty times. The alt text that every tool passed turns each cover link into a riddle. Because the image is the link, its text alternative is the link's name, and it must say where the link goes, for example the book's title (1.1.1 and 2.4.4).6 My chapter on images, video, colour and motion covers the rules.

On a product page he presses "Add to basket". The basket count in the header changes from 0 to 1. NVDA says nothing. A customer who cannot see the header has no idea whether it worked. WCAG 2.1 success criterion 4.1.3, Status Messages, asks that such a message be announced without moving focus, which a polite live region does.6 The same silence follows when he applies a filter on search results: the list changes, nothing is said.

He picks up his iPhone, turns on VoiceOver and goes through the checkout. He leaves the email field empty and submits. A red line of text appears under the field, but VoiceOver reads nothing, and when he swipes back to the field it is announced without the error. The message is not tied to the field, a failure of 3.3.1.6 My chapter on an accessible checkout walks through that fix.

Four more issues. None of them was visible to a tool, and none of them was visible to Ade's eyes either. He had to listen.

The result of the hour

At the end of the hour Ade writes up his list for Sinéad. He sorts it by one question: does this stop a customer from finding, choosing or paying for a book? Anything on that path, the shop's critical path, goes first. If the idea of a critical path is new, my article on the critical path in software explains how to find yours.

Bar chart of an invented example. Of 18 distinct issues found in a one hour test of an online bookshop: automated tools found 8, the keyboard pass found 3, zoom and reflow found 2, contrast checked by hand found 1, and the screen reader pass found 4. Ten of the 18 issues, more than half, were found only by a person. Tap or click to open full screen and move around

Example: Ade's write up for Sinéad

Ade's note to Sinéad fits on one page:

  • Blocks a sale, fix this week: delivery options unreachable by keyboard, checkout errors not announced, Eircode field hidden under the sticky header at 200%, no visible focus.
  • Makes buying hard, fix this month: cover links all called "book cover", basket and filter changes not announced, quick view dialog does not move focus, search results scroll sideways at 320 pixels.
  • Already fixed: the eight automated findings.
  • Not yet tested: real customers. "An hour at my desk is not the same as a morning with someone who uses this every day."

He adds one honest line: he tested five pages, and the account area, gift cards and events pages are still unknown.

Testing with disabled people: the test no checklist replaces

Everything so far was a developer imitating a user. The last layer is the real thing: people who use assistive technology every day, doing a real task on your site while you watch and listen. They find what passes every check and is still slow, tiring or confusing.

The W3C recommends involving users with disabilities throughout a project rather than only at the end, because informal evaluations during development are more effective than formal testing at the end.26 It also warns against over reading the result: what a couple of people find cannot be generalised to everyone with similar disabilities, or to people with other disabilities.26 A session is evidence, not a certificate.

The GOV.UK Service Manual has practical advice for running sessions.27 Ask participants to bring their own assistive technology, because their personal settings are hard to recreate on another device, and if they cannot bring it, visit them instead. Make sure the consent form itself is accessible. Talk to the participant directly, not to their helper or interpreter. Do not guess what they can or cannot do, ask. If you are not sure how to refer to their disability, ask what terminology they prefer.

My own advice on top: pay people for their time and expertise, as you would pay any specialist. You are asking for a skill, and a fair fee is part of treating it as one.

Example: three customers, one morning

Sinéad asks in the shop and on her newsletter whether any customers would test the website for a fee. Three people sign up, and Ade runs a 45 minute session with each in the back room of the shop, with the task "find a book you would like, and buy it as a gift".

Declan is blind and uses NVDA on his laptop and VoiceOver on his iPhone. He brings both. By now the cover links read the book titles and the basket announces itself. He finds something new: in the checkout, the explanation of "collect from the shop" is read out after the payment button, because the layout places it visually beside the options but the HTML puts it at the end of the page. He only learns about the free collection option after he has paid for delivery. That is a failure of 1.3.2, Meaningful Sequence.6

Máire has navigated by keyboard since a stroke two years ago. Focus is now visible and the delivery options work. What wears her down is the header: 48 links in the category menu come before the content, on every page, and there is no skip link. Getting from the top of a product page to "Add to basket" takes 52 presses of Tab. axe had passed the page because it has a proper <main> landmark, which a screen reader can jump to, but a keyboard has no such jump. A "Skip to main content" link as the first item fixes it (2.4.1).6

Bríd has low vision and reads at 200% zoom with high contrast turned on in Windows. She adds a book to the basket and a small confirmation appears in the corner, then vanishes after three seconds, long before she has panned to it. She asks Sinéad, reasonably, whether the book is in her basket or not. Ade makes the message stay until it is dismissed, and repeats the confirmation on the basket page.

Three people, one morning, three problems that no tool and no developer had found.

Put the automated checks in your test suite

Running tools by hand once is a snapshot. The site changes every week, and a new banner or a redesigned button can bring back the problems you fixed. The answer is to run the automated checks inside your automated tests, so a regression fails the build like any other broken test.

Playwright, a browser testing framework from Microsoft, documents exactly this with the @axe-core/playwright package.28 Its documentation also carries a fair warning: automated tests can detect some common problems, but many can only be discovered through manual testing, and it recommends a combination of automated testing, manual assessment and inclusive user testing.28

Two details from the documentation matter in practice. First, axe scans the page in the state it is in when you call it, so for a menu or a dialog, open it first, wait for it, then scan. Second, you can limit the scan to rules tagged for WCAG A and AA with the tags wcag2a, wcag2aa, wcag21a and wcag21aa.28

Example: Ade's accessibility tests for the critical path

Ade adds one test file to the shop. It scans the four pages of the buying journey, scans the quick view dialog once it is open, and checks that the first press of Tab lands on the new skip link.

// tests/accessibility.spec.js
import { test, expect } from '@playwright/test';
import AxeBuilder from '@axe-core/playwright';

const WCAG_AA = ['wcag2a', 'wcag2aa', 'wcag21a', 'wcag21aa'];
const criticalPages = ['/', '/search?q=galway', '/books/the-salthill-tide', '/basket'];

for (const path of criticalPages) {
  test(`no detectable WCAG A or AA violations on ${path}`, async ({ page }) => {
    await page.goto(path);
    const results = await new AxeBuilder({ page }).withTags(WCAG_AA).analyze();
    expect(results.violations).toEqual([]);
  });
}

test('the quick view dialog has no detectable violations once open', async ({ page }) => {
  await page.goto('/books/the-salthill-tide');
  await page.getByRole('button', { name: 'Quick view' }).click();
  await page.getByRole('dialog').waitFor();
  const results = await new AxeBuilder({ page })
    .include('[role="dialog"]')
    .withTags(WCAG_AA)
    .analyze();
  expect(results.violations).toEqual([]);
});

test('the first Tab lands on the skip link', async ({ page }) => {
  await page.goto('/');
  await page.keyboard.press('Tab');
  await expect(page.getByRole('link', { name: 'Skip to main content' })).toBeFocused();
});

Line by line: AxeBuilder wraps the axe engine for Playwright. page.goto(path) opens each page from the base address in the Playwright configuration. withTags(WCAG_AA) keeps only the rules for WCAG 2.0 and 2.1 at Levels A and AA. analyze() runs the scan, and expect(results.violations).toEqual([]) fails the test on a single violation. getByRole('button', { name: 'Quick view' }) finds the button the way a screen reader would, by role and accessible name, a small check in itself. include('[role="dialog"]') scans only the dialog. The last test presses Tab once and asserts that focus is on the skip link, so Máire's fix cannot quietly disappear.

The documentation also shows exclude() and disableRules() for known issues you cannot fix yet.28 Use them sparingly, write down why and until when, and never use them to make a red build green without a plan. A skipped rule is a known gap, and your accessibility statement should say so. If you want to see where tests like these sit next to your other tests, my article on unit, integration and end to end tests explains the levels.

How often to test: every release that touches the critical path

Every change to a template, a component or the checkout can break accessibility again. The law reflects this. In Ireland, S.I. No. 636 of 2023 carries the European Accessibility Act into Irish law, and it requires a service provider to "ensure that procedures are in place so that the provision of a service remains in conformity with the applicable accessibility requirements".29 The Act itself puts the obligations on the service provider, the business, not on its software vendor or plugin.30 A testing routine is exactly such a procedure. Alongside it, the Equal Status Acts count a failure by a service provider to do all that is reasonable to accommodate the needs of a person with a disability as discrimination, and that applies to your website as it is today, not as it was at launch.3132 My article on accessibility law for online businesses covers who must comply and from when.

A sensible rhythm for a small shop looks like this:

  • Every build: the automated checks in the test suite, on the critical path at least.
  • Every release that touches the critical path or a shared template: a manual keyboard, zoom and reflow pass on the changed journey, 15 to 20 minutes, and a screen reader pass on whatever changed.
  • Every quarter: the full hour, on every page type, not only the critical path.
  • Once a year, and before any redesign: sessions with people who use assistive technology.
Tap or click to open full screen and move around

Example: two releases at Claddagh Books

In March Ade changes the wording on the "About the shop" page. The automated checks run and pass. The change touches neither the buying journey nor a shared template, so he releases it and notes "automated only" in the release log.

In April Sinéad wants a new "gift wrap" option in the checkout. The automated checks pass. The change touches the critical path, so Ade does the manual pass. The new option is a custom toggle switch, and it cannot be operated with Space. He replaces it with a native checkbox styled to look the same, runs the screen reader pass, hears "Gift wrap, two euro fifty, checkbox, not checked", and releases. The log records what was tested, by whom and when. That log is also the backbone of an honest accessibility statement.

How I build and test to this

I build shops and SaaS products to WCAG 2.1 AA, and I say "built to WCAG 2.1 AA", never "guaranteed compliant". The testing is the mix in this article: automated checks in the test suite on every build, keyboard only and screen reader passes by hand on the key journeys, zoom and reflow checks at 200% and 320 pixels, sessions with people who use assistive technology where it is warranted, and a record of what was tested and fixed, because that record is what an independent audit builds on.

My own website has an accessibility panel built into the site's own code, with text size, high contrast and reduced motion. It is a set of preferences inside a site that is already built accessibly, not the thing that makes it accessible, and nothing like an overlay that claims to fix a site from the outside. Every chart on my site has a text description and works in light and dark mode.

Conclusion: tools first, people always

Run the free tools, fix what they find, and do not mistake a perfect score for an accessible site. The tools found eight of Ade's eighteen issues. The other ten needed a keyboard, a zoom level, a contrast checker and a pair of ears, and three more needed Declan, Máire and Bríd. Put the automated checks in your test suite so the basics never come back, do a short manual pass every time the buying journey changes, and invite real customers at least once a year. Then write down what you tested. That record is your answer to Sinéad's question, and to anyone else who asks.

Want to know what your shop's checkout sounds like to a screen reader, and what the tools have missed? Book the free workshop and we will test the critical path together, and the first two weeks of work are on me.

Sources

  1. 1.

    W3C Web Accessibility Initiative, Evaluating Web Accessibility Overview, W3C, https://www.w3.org/WAI/test-evaluate/

  2. 2.

    Deque Labs, axe core: accessibility engine for automated web UI testing, GitHub repository and documentation, https://github.com/dequelabs/axe-core

  3. 3.

    Google, Lighthouse accessibility score, Chrome for Developers documentation, https://developer.chrome.com/docs/lighthouse/accessibility/scoring

  4. 4.

    WebAIM, WAVE Help, Utah State University, https://wave.webaim.org/help

  5. 5.

    W3C Web Accessibility Initiative, Selecting Web Accessibility Evaluation Tools, W3C, https://www.w3.org/WAI/test-evaluate/tools/selecting/

  6. 6.

    W3C, Web Content Accessibility Guidelines (WCAG) 2.1, W3C Recommendation, 5 June 2018, updated 6 May 2025, success criteria 1.1.1, 1.3.2, 1.4.3, 1.4.11, 2.1.1, 2.1.2, 2.4.1, 2.4.3, 2.4.4, 2.4.6, 3.1.1, 3.3.1, 3.3.2, 3.3.3 and 4.1.3, https://www.w3.org/TR/WCAG21/

  7. 7.

    Mehmet Duran, What we found when we tested tools on the world's least accessible webpage, Accessibility in government blog, Government Digital Service, 24 February 2017, https://accessibility.blog.gov.uk/2017/02/24/what-we-found-when-we-tested-tools-on-the-worlds-least-accessible-webpage/

  8. 8.

    Deque Systems, Deque study shows its automated testing identifies 57% of digital accessibility issues, surpassing accepted industry benchmarks, press release, 10 March 2021, https://www.deque.com/blog/automated-testing-study-identifies-57-percent-of-digital-accessibility-issues/

  9. 9.

    Deque Systems, The Automated Accessibility Coverage Report, deque.com, https://www.deque.com/automated-accessibility-testing-coverage/

  10. 10.

    WebAIM, The WebAIM Million: the 2026 report on the accessibility of the top 1,000,000 home pages, Utah State University, 2026, https://webaim.org/projects/million/

  11. 11.

    W3C Web Accessibility Initiative, Understanding Success Criterion 2.4.7: Focus Visible, https://www.w3.org/WAI/WCAG21/Understanding/focus-visible.html

  12. 12.

    W3C Web Accessibility Initiative, Understanding Success Criterion 1.4.4: Resize Text, https://www.w3.org/WAI/WCAG21/Understanding/resize-text.html

  13. 13.

    W3C Web Accessibility Initiative, Understanding Success Criterion 1.4.10: Reflow, https://www.w3.org/WAI/WCAG21/Understanding/reflow.html

  14. 14.

    W3C Web Accessibility Initiative, Understanding Success Criterion 1.4.3: Contrast (Minimum), https://www.w3.org/WAI/WCAG21/Understanding/contrast-minimum.html

  15. 15.

    Government Digital Service, Testing for accessibility, GOV.UK Service Manual, https://www.gov.uk/service-manual/helping-people-to-use-your-service/testing-for-accessibility

  16. 16.

    W3C, Web Content Accessibility Guidelines (WCAG) 2.2, W3C Recommendation, success criterion 2.4.11 Focus Not Obscured (Minimum), Level AA, https://www.w3.org/TR/WCAG22/

  17. 17.

    WebAIM, Screen Reader User Survey #10 Results, survey of December 2023 and January 2024, 1,539 valid responses, updated 22 February 2024, https://webaim.org/projects/screenreadersurvey10/

  18. 18.

    Government Digital Service, Testing with assistive technologies, GOV.UK Service Manual, https://www.gov.uk/service-manual/technology/testing-with-assistive-technologies

  19. 19.

    NV Access, NVDA User Guide, quick start and browse mode commands, https://download.nvaccess.org/documentation/userGuide.html

  20. 20.

    Apple, Turn VoiceOver on or off on Mac, VoiceOver User Guide, https://support.apple.com/guide/voiceover/turn-voiceover-on-or-off-vo2682/mac

  21. 21.

    Apple, Use the VoiceOver rotor on Mac, VoiceOver User Guide, https://support.apple.com/guide/voiceover/use-the-voiceover-rotor-mchlp2719/mac

  22. 22.

    Apple, Turn on and practise VoiceOver on iPhone, iPhone User Guide, https://support.apple.com/guide/iphone/turn-on-and-practice-voiceover-iph3e2e415f/ios

  23. 23.

    Apple, Learn VoiceOver gestures on iPhone, iPhone User Guide, https://support.apple.com/guide/iphone/learn-voiceover-gestures-iph3e2e2281/ios

  24. 24.

    Google, Turn on TalkBack, Android Accessibility Help, https://support.google.com/accessibility/android/answer/6007100

  25. 25.

    Google, Use TalkBack gestures, Android Accessibility Help, https://support.google.com/accessibility/android/answer/6151827

  26. 26.

    W3C Web Accessibility Initiative, Involving Users in Evaluating Web Accessibility, W3C, https://www.w3.org/WAI/test-evaluate/involving-users/

  27. 27.

    Government Digital Service, Running research sessions with disabled people, GOV.UK Service Manual, https://www.gov.uk/service-manual/user-research/running-research-sessions-with-people-with-disabilities

  28. 28.

    Microsoft, Accessibility testing, Playwright documentation, https://playwright.dev/docs/accessibility-testing

  29. 29.

    Minister for Children, Equality, Disability, Integration and Youth, S.I. No. 636/2023, European Union (Accessibility Requirements of Products and Services) Regulations 2023, regulation 14(3), electronic Irish Statute Book, 2023, https://www.irishstatutebook.ie/eli/2023/si/636/made/en/print

  30. 30.

    European Parliament and Council, Directive (EU) 2019/882 of 17 April 2019 on the accessibility requirements for products and services (European Accessibility Act), Article 13, Obligations of service providers, EUR Lex, 2019, https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=CELEX:32019L0882

  31. 31.

    Oireachtas, Equal Status Act 2000, section 4, Discrimination on ground of disability, electronic Irish Statute Book, https://www.irishstatutebook.ie/eli/2000/act/8/section/4/enacted/en/html

  32. 32.

    Oireachtas, Equal Status Act 2000, section 5, Disposal of goods and provision of services, electronic Irish Statute Book, https://www.irishstatutebook.ie/eli/2000/act/8/section/5/enacted/en/html