An accessible checkout: the critical path for every customer

TL;DR

Difficulty: Intermediate

The checkout is your critical path, and a disabled customer who cannot pass one step leaves without telling you: in a UK survey only 8% of disabled shoppers who struggle ever contact the site.

  • Offer guest checkout first, and never block pasting codes or passwords (WCAG 2.2, 3.3.8).
  • Announce basket changes with a status region (4.1.3), and keep address fields typeable when the lookup fails (3.3.7).
  • Delivery options are one radio group with a legend, and choosing one never submits the form (1.3.1, 3.2.2).
  • The payment provider owns its frame and the bank owns its check, but you own the page around them, how you configure them, and a second way to pay.
  • Warn before time runs out (2.2.1), put the order number in the confirmation title (2.4.2), and test the whole path by keyboard, screen reader and an automated end to end test on every release.

A customer who cannot get through your checkout rarely complains. They leave, and the abandoned basket looks like a change of mind. This chapter walks the checkout of an invented Brighton bookshop step by step with a blind customer, shows what stopped him at each step and the HTML that fixed it, and explains what your payment provider and the customer's bank answer for, and what stays with you.

Can every customer actually pay you?

Most shop owners never hear about the customer who could not pay. There is no error in the logs and no angry email. There is only a basket that was filled and then left, and in the analytics it looks exactly like someone who changed their mind.

For disabled customers that silence is the norm. In the Click Away Pound survey of 2019, 69% of disabled online shoppers with access needs said they simply leave a website they find hard to use, and only 8% of those who have difficulty ever contact the site owner. The authors put the spending power of the people who click away at £17.1 billion a year in the UK alone1. And the barriers are common: the 2026 WebAIM Million, an automated scan of the home pages of the top million websites, found missing form labels on 51% of them, and a third of all form inputs it detected had no proper label2.

This chapter walks the checkout one step at a time, from the basket to the confirmation. For each step it shows what blocks people, the fix in plain HTML, and who is responsible when part of the step belongs to someone else, such as your payment provider or the customer's bank. Labels, errors and help text are covered in the chapter on accessible forms, so I build on it rather than repeat it.

Example: a Saturday morning basket at Brightwater Books

Brightwater Books is an invented independent bookshop in Brighton with an online shop at brightwaterbooks.example. Ruth owns it. Ade is the developer who looks after the site. This series follows them.

Tom is a customer, also invented. He is blind and uses NVDA, a free screen reader for Windows3. A screen reader turns the page into speech, and Tom moves around with the keyboard: Tab between links, buttons and fields, arrow keys to read, Enter and Space to press. He wants to buy two copies of a local walking guide, one for him and one for his sister.

On Saturday morning Tom gets as far as the delivery step, cannot find a way to choose a delivery option, and gives up. He buys the book somewhere else. Ruth sees an abandoned basket worth £25.98. Nobody tells her why. The rest of this chapter is the list of things Ade finds when he walks Tom's route himself, and what he changes.

The checkout is your critical path, so it gets the strictest standard

A critical path is the shortest chain of steps that must work for the business to earn money. I explain the idea in full in the critical path in software. In a shop, the checkout is that path. A broken blog page costs you a reader. A broken checkout costs you the sale.

The accessibility standard recognises this too. The Web Content Accessibility Guidelines (WCAG) apply to every page, but one criterion is written specifically for pages where people commit money: success criterion 3.3.4, Error Prevention (Legal, Financial, Data), at Level AA. It says that when a page causes a financial transaction, the submission must be reversible, or checked for errors with a chance to correct them, or reviewable and confirmable before it is final4. The W3C's own explanation gives the example of an order page that shows the items, quantities, delivery address and payment method so the shopper can check them before confirming5.

The law points the same way. In the UK, the Equality Act 2010 forbids service providers from discriminating against disabled people and places on them a duty to make reasonable adjustments67. If you sell to customers in the European Union, the European Accessibility Act names online shops directly and requires that the identification, security and payment functions of an ecommerce service are perceivable, operable, understandable and robust (Annex I, Section IV, point (g))8. The benchmark the EU uses is EN 301 549, which builds on WCAG at Level AA9. I cover who must comply and from when in accessibility law for online businesses. For this chapter, one point matters: the duty sits with the business that runs the shop. Not with the platform, not with the payment provider, not with a plugin.

Example: mapping the checkout at Brightwater Books

Ade writes the checkout down as a list of steps, one line each, before touching any code. That list is the critical path he will test:

  1. Basket: change quantities, see the total.
  2. Sign in or check out as a guest.
  3. Delivery address, with a postcode lookup.
  4. Delivery option: standard, next day or collect from the shop.
  5. Review the order.
  6. Payment: card details in the payment provider's component.
  7. The bank's security check, when the bank asks for one.
  8. Order confirmation.

Then, for each step, he asks one question: can a customer who uses only a keyboard and a screen reader complete it, and know that they have? If the answer is no anywhere, the customer leaves at that step, and everything after it is irrelevant. That is why the diagram below is a chain of gates rather than a list of nice to haves.

Tap or click to open full screen and move around

Let people buy without making an account

Forcing an account is a barrier for everyone, and a bigger one for some. In Baymard Institute's survey of US online shoppers, 18% named "the site wanted me to create an account" as a reason they abandoned an order, and 17% named a checkout that was too long or complicated10. An account adds a password to invent, often a code to copy from an email or a text message, and one more form with its own errors.

For disabled customers the cost rises. WCAG 2.2 added success criterion 3.3.8, Accessible Authentication (Minimum), at Level AA. It says that no step of signing in may require a cognitive function test, such as remembering a password or solving a puzzle, unless there is an alternative or a mechanism to help11. The W3C's explanation is concrete: let password managers fill the fields, never block pasting, and do not make people transcribe a one time code by hand. A code field split into one box per digit that refuses a pasted code is called out as a problem12.

The fix that removes most of this is simple: offer guest checkout as an equal choice, and offer to create an account after the order is placed, with the details already filled in.

Example: guest first at Brightwater Books

Brightwater's old step two showed a sign in form with "Create an account to continue" underneath. New customers had to register, confirm a six digit code that arrived by email, and come back. The code field was six separate boxes and pasting into it did nothing.

Ade changes three things. First, the step now opens with a plain question, written as a radio group: "How would you like to check out?" with "As a guest" and "Sign in to my account" as the two options, guest first. Second, the sign in form uses proper labels and autocomplete="email" and autocomplete="current-password", so Tom's password manager can fill it. Third, the code field becomes a single input with autocomplete="one-time-code" and inputmode="numeric", which accepts a pasted code. On the confirmation page, a guest now sees "Save your details for next time" with only a password to choose.

Tom checks out as a guest. He never sees a code at all.

Basket changes must be announced

When a customer changes a quantity, the page usually updates the line total and the basket total without reloading. A sighted customer sees the numbers change. A screen reader user hears nothing, because nothing told the screen reader that anything happened.

WCAG 2.1 covers this with success criterion 4.1.3, Status Messages, at Level AA: a status message must be exposed to assistive technology through a role or property, so it can be announced without moving the user's focus4. The W3C's explanation uses exactly this case: after "Add to Shopping Cart", a message such as "5 items" should be announced as "Shopping cart, five items"13.

The tool for this is a live region: an element that the screen reader watches and reads aloud when its text changes. An element with role="status" is a polite live region, which means the screen reader waits until the user pauses before speaking, and reads the whole message rather than only the changed part14. One detail decides whether it works at all: the live region must already be in the page, empty, before you put text into it. MDN, the browser makers' reference documentation, says the most reliable way is to include it in the initial markup15.

Example: the quantity buttons in Brightwater's basket

Here is what Ade found. The minus and plus are styled div elements with click handlers, the number is a plain span, and the totals update silently:

<!-- Before: not reachable by keyboard, not named, updates silently -->
<div class="qty">
  <div class="qty-minus" onclick="changeQty('chalk', -1)">&minus;</div>
  <span id="qty-chalk">1</span>
  <div class="qty-plus" onclick="changeQty('chalk', 1)">+</div>
</div>
<p>Total: <span id="basket-total">£12.99</span></p>

Tom cannot reach the div elements with Tab at all, because a div is not focusable. If he could, his screen reader would announce "plus" with no word about which book. And the new total would never be spoken.

Here is the fix:

<!-- After: a labelled field, a real button, and a status region -->
<form action="/basket/lines/chalk" method="post">
  <label for="qty-chalk">Quantity of Chalk Paths of the South Downs</label>
  <input id="qty-chalk" name="quantity" type="number"
         min="0" max="20" value="1" inputmode="numeric">
  <button type="submit">Update basket</button>
</form>
<p>Total: <span id="basket-total">£12.99</span></p>

<!-- Present and empty when the page loads -->
<p role="status" id="basket-status"></p>
// After the server confirms the change, say what changed in one sentence
function announceBasket(items, total) {
  document.getElementById('basket-total').textContent = total;
  document.getElementById('basket-status').textContent =
    `Basket updated: ${items} items, total ${total}.`;
}

Attribute by attribute, for a beginner. The label with for="qty-chalk" is tied to the input with the same id, so the screen reader announces "Quantity of Chalk Paths of the South Downs" instead of just "edit". type="number" with min and max lets the arrow keys change the value, and inputmode="numeric" brings up the number keypad on phones. The button is focusable and works with Enter and Space without extra code. The form posts to the server, so it still works if the JavaScript fails. And role="status" makes the empty paragraph a polite live region, so Tom hears "Basket updated: 2 items, total £25.98" once he pauses. The book, by the way, is invented too.

Address lookup that also works by hand

A postcode lookup saves typing: enter the postcode, choose your address from a list. When it works, everyone gains. When it does not, it can be a dead end. New build homes are often missing from address data, flats and rural addresses come out wrong, and a lookup built from a custom dropdown can be impossible to use without a mouse.

The GOV.UK Design System, the UK government's public pattern library, is clear on this: a lookup must make plain that it only works for UK addresses, and there must always be a manual option16. It also recommends the autocomplete attribute on each address field, which lets the browser fill the address and helps meet WCAG success criterion 1.3.5, Identify Input Purpose164.

The second rule of this step is about not asking twice. WCAG 2.2 added success criterion 3.3.7, Redundant Entry, at Level A: information a customer already entered in the same process must be filled in automatically or offered for selection11. The W3C's own example is a checkbox that says the billing address is the same as the delivery address17.

Example: Tom's new flat

Tom moved into a new build flat last year. Brightwater's lookup returned "No addresses found" and showed nothing else. The address fields only appeared after a successful lookup, so there was no way forward.

Ade rebuilds the step so the fields are always there, and the lookup only fills them in:

<fieldset>
  <legend>Delivery address</legend>

  <label for="postcode">Postcode</label>
  <input id="postcode" name="postcode" autocomplete="postal-code">
  <button type="button" id="find-address">Find my address</button>
  <p role="status" id="lookup-status"></p>

  <label for="address-line-1">Address line 1</label>
  <input id="address-line-1" name="address_line_1" autocomplete="address-line1">

  <label for="address-line-2">Address line 2 (optional)</label>
  <input id="address-line-2" name="address_line_2" autocomplete="address-line2">

  <label for="town">Town or city</label>
  <input id="town" name="town" autocomplete="address-level2">
</fieldset>

<input type="checkbox" id="billing-same" name="billing_same" checked>
<label for="billing-same">My billing address is the same as my delivery address</label>

The fieldset and legend group the fields under one name, so Tom hears "Delivery address, group" when he enters it. Each autocomplete value tells the browser exactly what the field is for, so it can offer Tom's saved address in one go. When the lookup succeeds, the status region says "4 addresses found. Choose yours from the list that follows, or type it into the address fields." When it fails, it says "No addresses found for BN1 4AB. Please type your address in the fields that follow." Either way, Tom knows what happened and what to do. And the billing checkbox, ticked by default, means he never types the address a second time.

Delivery options are one radio group with a name

Delivery is a choice of exactly one option from a few, and HTML has a control built for that: radio buttons, grouped in a fieldset with a legend. The W3C's forms tutorial says radio button groups should always be grouped with fieldset18. That grouping is what WCAG success criterion 1.3.1, Info and Relationships, asks for: the structure a sighted customer sees must be available to assistive technology too4.

Built properly, a radio group gives the keyboard user a lot for free. Tab moves into the group, the arrow keys move between options, and the screen reader announces the group's name, the option's name, and its position, such as "2 of 3".

Two details matter. Put the price and the timing inside each option's label, so they are read together with the option. And do not let choosing an option submit the form or jump to another page. That would break success criterion 3.2.2, On Input, which says changing a setting must not cause a change of context unless the user was told beforehand419. A keyboard user moving through the options with the arrow keys selects each one on the way, so an option that submits on change sends them somewhere they did not ask to go.

Example: Brightwater's delivery step, in HTML and in Rails

The old delivery step was three styled boxes that changed colour when clicked. Tom's screen reader read them as plain text, and Tab skipped them. This is where he gave up.

Ade's fixed version in plain HTML:

<form action="/checkout/delivery" method="post">
  <fieldset>
    <legend><h2>How would you like your books delivered?</h2></legend>

    <div>
      <input type="radio" id="delivery-standard" name="delivery" value="standard" checked>
      <label for="delivery-standard">Standard delivery, £3.50, 3 to 5 working days</label>
    </div>
    <div>
      <input type="radio" id="delivery-next-day" name="delivery" value="next_day">
      <label for="delivery-next-day">Next day delivery, £6.95, order by 2pm</label>
    </div>
    <div>
      <input type="radio" id="delivery-collect" name="delivery" value="collect">
      <label for="delivery-collect">Collect from the shop in Brighton, free, ready in 2 hours</label>
    </div>
  </fieldset>

  <button type="submit">Continue to review your order</button>
</form>

All three inputs share name="delivery", which is what makes them one group where only one can be chosen. The legend names the group, and a heading inside the legend lets Tom also find the step with his screen reader's heading navigation. The submit button says exactly what happens next.

Brightwater's shop runs on Ruby on Rails, so Ade writes the same thing as a view with the Rails form helpers:

<%= form_with model: @order, url: checkout_delivery_path, method: :patch do |form| %>
  <fieldset>
    <legend><h2>How would you like your books delivered?</h2></legend>

    <% @delivery_options.each do |option| %>
      <div>
        <%= form.radio_button :delivery_option, option.code %>
        <%= form.label :delivery_option, value: option.code do %>
          <%= option.name %>, <%= number_to_currency(option.price, unit: "£") %>, <%= option.timing %>
        <% end %>
      </div>
    <% end %>
  </fieldset>

  <%= form.submit "Continue to review your order" %>
<% end %>

form.radio_button :delivery_option, option.code creates an input with an id such as order_delivery_option_standard. form.label :delivery_option, value: option.code creates a label whose for points at exactly that id, so each label belongs to its own radio button. The loop means a new delivery option added in the admin area appears with a correct label automatically, and nobody has to remember the accessibility each time.

Payment fields: what the provider owns and what you own

Card details are usually typed into a component supplied by the payment provider. It sits inside an iframe, a small window from the provider's own website embedded in your page. That is good for security, because the card number never touches your server. It also means part of your checkout is code you cannot edit.

So who is responsible for what? Start with what the provider documents. Adyen, for example, publishes an Accessibility Conformance Report (ACR, a vendor's statement of how a product meets each WCAG criterion) for its Web Drop in and Components. It states that they conform to WCAG 2.2 at Level AA from version 6.40.2, and it is explicit about the limits: the report covers only the prebuilt interface in its default rendering, not an interface you build yourself on their API, and customising it, for example by changing colours, can lower contrast below the required level20. The conformance report for a WooCommerce extension that uses Stripe's hosted payment fields makes the same point from the other side: parts of that checkout are controlled by Stripe, not by the shop software around them21.

That gives a fair split. The provider answers for the inside of its frame, as delivered. You answer for everything around it and for how you configure it: the heading that says what this step is, the version you run, the colours you set, your error messages, where focus goes after a failed payment, and a second way to pay. From the customer's point of view there is no split at all. It is your shop.

Tap or click to open full screen and move around

Example: the payment step around the frame

Brightwater's payment step had a provider component dropped into an otherwise empty page. No heading, no text, and when a card was declined, a red box appeared at the top of the page that no screen reader announced. Ade had also set the component's placeholder text to a pale grey to match the brand, which pushed it below the contrast the provider shipped with.

Ade's new frame around the component:

<section aria-labelledby="pay-heading">
  <h2 id="pay-heading">Pay by card</h2>
  <p>Your card details go straight to our payment provider over a secure connection.
     Brightwater Books never sees or stores them.</p>

  <div id="card-component"><!-- the provider's component is mounted here --></div>

  <!-- Present and empty when the page loads -->
  <div id="payment-error" role="alert"></div>

  <button type="submit" id="pay-button">Pay £29.48</button>
  <p><a href="/checkout/payment/other">Pay another way</a></p>
</section>
// When the provider reports a declined payment
function showPaymentError(message) {
  const box = document.getElementById('payment-error');
  box.textContent = message; // e.g. "Your card was declined. You have not been charged. Try another card or pay another way."
  document.getElementById('pay-button').focus();
}

The section with aria-labelledby becomes a named region, "Pay by card", that Tom can jump to, and the paragraph tells him what the embedded component is before he enters it. role="alert" is an assertive live region, so a declined payment is spoken at once, and focus returns to the Pay button, ready for another try. The button names the amount, so there is no surprise at the moment of commitment. "Pay another way" leads to a second method, such as a wallet or a bank transfer, so a barrier inside one component is never the end of the road. And Ade put the placeholder colour back to the provider's default, because a customisation that lowers contrast is his to fix, not the provider's.

The bank's check is the bank's screen, and the frame around it is yours

In the UK and across Europe, many card payments need strong customer authentication: the customer proves it is really them, usually through 3D Secure, the card networks' protocol for it. Stripe's documentation explains that the requirement comes from the card issuer, the customer's bank, which may ask for a password, a one time code sent to a phone, or confirmation in its banking app22.

The important part for accessibility: the bank controls the screen. Stripe's documentation says plainly that you cannot customise the authentication interface on the web to match your site, because the bank that issued the card controls its fonts and colours. What you can choose is how and where it appears: in a modal dialog above your payment page, inline in the page, or as a redirect to the bank's own site22.

So the inside of that screen is the bank's responsibility, and the way it is opened is yours. If it appears in a layer that keyboard focus never enters, Tom is stuck behind it. If it opens in a modal dialog that takes focus and gives it back, he can complete it.

The native HTML dialog element does most of that work. When it is opened with showModal(), the browser moves focus to the first focusable element inside, makes the rest of the page inert so focus cannot wander behind it, and lets the Escape key close it23.

Example: Tom's bank asks for a code

Tom's bank asks for a code sent to his phone. On the old site, the bank's frame appeared over the page in a styled box, but focus stayed on the Pay button underneath. Tom pressed Tab and wandered through the footer.

Ade places the bank's frame inside a dialog:

<dialog id="bank-check" aria-labelledby="bank-check-heading">
  <h2 id="bank-check-heading" tabindex="-1" autofocus>Your bank wants to confirm it's you</h2>
  <p>This check is run by your bank. When it is done, you will come back here automatically.</p>
  <div id="bank-frame"><!-- the provider places the bank's frame here --></div>
</dialog>
const dialog = document.getElementById('bank-check');

function openBankCheck() {
  dialog.showModal(); // focus moves to the heading inside, the page behind goes inert
}

function bankCheckFinished(result) {
  dialog.close();
  if (result === 'succeeded') {
    window.location = '/checkout/confirmation';
  } else {
    showPaymentError('Your bank could not confirm the payment. You have not been charged. Try again or pay another way.');
  }
}

The heading with autofocus and tabindex="-1" receives focus when the dialog opens, so Tom hears "Your bank wants to confirm it's you, dialog" and knows where he is. The frame's content then belongs to his bank. When the check ends, either he lands on the confirmation page, or the dialog closes and the error is announced with focus back on the Pay button. If a bank's own screen turns out to be unusable for a customer, Ruth cannot fix it, but she can make sure "Pay another way" is always one step away.

Time limits in the payment step

Checkouts are full of clocks. A basket may be held for a fixed time, a payment session expires, a bank check runs out. Someone using a screen reader, someone who types with a head pointer, someone who needs to stop and rest, all take longer than the person who set the timer.

WCAG success criterion 2.2.1, Timing Adjustable, is at Level A, the basic floor. For each time limit you set, the customer must be able to turn it off, or adjust it to at least ten times the default, or be warned before it runs out and given at least 20 seconds to extend it with a simple action, at least ten times. Real time events, essential limits and limits over 20 hours are exempt24. A limit inside the bank's check is the bank's, but the ones you set are yours. Stripe's documentation for its mobile SDKs, for instance, requires the authentication timeout to be at least five minutes22.

Example: the ten minute basket hold

Brightwater held signed first editions in a basket for ten minutes, then released them silently. Tom lost one that way while he was working through the address step.

Ade keeps the hold, because signed copies are scarce, but makes it visible and extendable. Two minutes before the hold ends, a dialog opens:

<dialog id="hold-warning" aria-labelledby="hold-heading">
  <h2 id="hold-heading">Your basket is held for 2 more minutes</h2>
  <p>Signed copies are held for you while you check out. Do you need more time?</p>
  <form method="dialog">
    <button value="extend" autofocus>Hold my basket for another 20 minutes</button>
    <button value="leave">No, I'm finished</button>
  </form>
</dialog>

method="dialog" on the form closes the dialog when either button is pressed and passes the button's value back to the page, which then calls the server to extend the hold. Focus lands on the extend button, so a single press of Enter is the simple action the criterion asks for. Two minutes is far more than the 20 seconds required. And the server lets the hold be extended as many times as needed, which covers the "at least ten times" condition with room to spare.

Review before paying, and errors that say what to do

Success criterion 3.3.4 asks that a financial transaction is reversible, checked, or confirmable4. The usual way to meet it in a shop is a review step: one page that lists the books, quantities, delivery option, address and total, each with a way to change it, followed by the payment step. It helps everyone, and it helps most the customer who cannot glance across a page to spot that the quantity says 20 instead of 2.

When something is wrong, the GOV.UK pattern is a good model: an error summary at the top of the main content, focus moved to it, every error linked to its field, and "Error:" added to the start of the page title so a screen reader announces it first25. The field by field detail is in the forms chapter. The checkout specific part is the round trip: a checkout that validates on the server and reloads the page must still name the problem in text, keep what the customer typed, and put focus on the summary.

Example: the review page catches a typo

Tom types "22" as the quantity by mistake. On the review page he hears the table read row by row: "Chalk Paths of the South Downs, quantity 22, £285.78". Each row has a "Change" link whose full name, for the screen reader, is "Change quantity of Chalk Paths of the South Downs", so he does not hear six identical "Change" links. He follows it, corrects the number, and the basket status says "Basket updated: 2 items, total £25.98." Without the review step he would have found out from his bank statement.

An order confirmation that is announced

The last step is the one that matters most to the customer: did it work? A confirmation page that says only "Thanks!" in large green letters answers that for a sighted customer at a glance. For a screen reader user, the first thing read on a new page is its title.

So the page title and the main heading must both carry the news, with the order number. The GOV.UK confirmation page pattern says a confirmation page must include the reference number, if there is one, and what happens next and when26. WCAG success criterion 2.4.2, Page Titled, at Level A, asks that every page has a title that describes its topic or purpose4.

When the confirmation is a full new page load, the screen reader reads the new title on its own. When the checkout is a single page application that swaps content without loading a new page, the title does not get read automatically, and you need to move focus to the new main heading yourself.

Example: the confirmation page Tom hears

<title>Order BB10482 confirmed, Brightwater Books</title>
...
<main>
  <h1 tabindex="-1" id="confirmation-heading">Thank you, Tom. Order BB10482 is confirmed.</h1>
  <p>A confirmation email is on its way to the address you gave.</p>
  <p>Your two books will arrive in 3 to 5 working days by standard delivery.</p>
  <p><a href="/orders/BB10482/receipt">Download your receipt as a PDF</a></p>
</main>
// Only for a checkout that changes the page without a full reload
document.title = 'Order BB10482 confirmed, Brightwater Books';
document.getElementById('confirmation-heading').focus();

The title starts with what matters, the order and its state, and ends with the shop's name. tabindex="-1" on the heading lets script move focus to it without adding it to the Tab order. Tom hears "Order BB10482 confirmed, Brightwater Books", then the heading, then what happens next. He knows he has paid, for what, and when it will arrive.

Test the whole path, every release

Each fix above can be undone by the next change: a new theme, a provider upgrade, a radio group replaced by prettier boxes. So the checkout is tested as one path, end to end, three ways.

By keyboard, with the mouse out of reach: from the basket to the confirmation, Tab, Shift and Tab, Enter, Space, the arrow keys in radio groups, and Escape in dialogs. You should always see where focus is, never get stuck, and never need the mouse. The keyboard chapter covers this routine in full.

With a screen reader: NVDA on Windows is free3, VoiceOver is built into every Mac and iPhone. Listen to each step. Is every field named? Is every change announced? Do you know at the end that you have paid? The testing chapter explains how to get started with each one and what tools can and cannot find.

As an automated end to end test that runs on every release. Playwright, a browser testing tool, documents how to run the axe accessibility engine inside a test with the @axe-core/playwright package, and says plainly that automated checks catch only some problems, so manual testing and testing with disabled people stay necessary27. How such a test fits next to unit and integration tests is in unit, integration and end to end tests, and the failures worth testing deliberately are in testing the unhappy path.

This is how I build and test too: shops built to WCAG 2.1 AA from the first commit, automated accessibility checks in the test suite on every build, keyboard and screen reader passes by hand on the key journeys, of which the checkout is the first, and sessions with people who use assistive technology where it is warranted.

Example: Ade's checkout test

Ade writes one Playwright test that walks Tom's route with the keyboard and checks each step with axe. It runs against the payment provider's test environment, never against live cards.

import { test, expect } from '@playwright/test';
import AxeBuilder from '@axe-core/playwright';

async function expectNoAxeViolations(page) {
  const results = await new AxeBuilder({ page })
    .withTags(['wcag2a', 'wcag2aa', 'wcag21a', 'wcag21aa'])
    .analyze();
  expect(results.violations).toEqual([]);
}

test('a guest can buy two books using only the keyboard', async ({ page }) => {
  await page.goto('/basket');
  await expectNoAxeViolations(page);

  // Basket: change the quantity and hear the new total
  await page.getByLabel('Quantity of Chalk Paths of the South Downs').fill('2');
  await page.getByRole('button', { name: 'Update basket' }).press('Enter');
  await expect(page.getByRole('status')).toHaveText('Basket updated: 2 items, total £25.98.');

  // Guest checkout, no account
  await page.getByRole('link', { name: 'Go to checkout' }).press('Enter');
  await page.getByRole('radio', { name: 'As a guest' }).focus();
  await page.keyboard.press('Space');
  await page.getByRole('button', { name: 'Continue to delivery address' }).press('Enter');

  // Address typed by hand after a failed lookup
  await page.getByLabel('Postcode').fill('BN1 4AB');
  await page.getByRole('button', { name: 'Find my address' }).press('Enter');
  await expect(page.locator('#lookup-status')).toContainText('No addresses found');
  await page.getByLabel('Address line 1').fill('Flat 3, 12 Example Road');
  await page.getByLabel('Town or city').fill('Brighton');
  await page.getByRole('button', { name: 'Continue to delivery options' }).press('Enter');

  // Delivery: arrow keys move through one radio group
  await page.getByRole('radio', { name: /Standard delivery/ }).focus();
  await page.keyboard.press('ArrowDown');
  await expect(page.getByRole('radio', { name: /Next day delivery/ })).toBeChecked();
  await page.keyboard.press('ArrowUp');
  await expectNoAxeViolations(page);
  await page.getByRole('button', { name: 'Continue to review your order' }).press('Enter');

  // Review, then pay with the provider's test card
  await expect(page.getByRole('heading', { name: 'Check your order' })).toBeVisible();
  await page.getByRole('button', { name: 'Continue to payment' }).press('Enter');
  await expectNoAxeViolations(page);
  // ...fill the provider's test card fields here, as its own testing guide describes

  // Confirmation: the title carries the news
  await expect(page).toHaveTitle(/Order BB\d+ confirmed, Brightwater Books/);
  await expect(page.getByRole('heading', { level: 1 })).toContainText('is confirmed');
});

Every step finds elements by role and accessible name, the way a screen reader does, with getByRole and getByLabel. If someone removes a label, the test cannot find the field and fails. That is the point: the test fails where Tom would have failed. The axe scans catch the mechanical problems, such as a missing label or low contrast, on each page. What the test cannot tell Ade is whether the sentences make sense when spoken, which is why he still listens to the path with NVDA before each release that touches the checkout.

Worked example: Tom's checkout, before and after

Here is the whole route as Tom met it on Saturday and as he meets it after Ade's fixes, in the order of the map Ade drew at the start. Everything is invented.

StepBeforeAfterWCAG
BasketPlus and minus were div elements, unreachable by Tab, total changed in silenceLabelled quantity field, Update basket button, status region2.1.1, 4.1.2, 4.1.3
Sign inAccount required, six box code that refused pastingGuest checkout first, account offered after the order3.3.8
AddressLookup found nothing, fields never appearedFields always present, result announced, billing reused1.3.5, 3.3.7
DeliveryStyled boxes that Tab skipped: where Tom leftOne radio group with a legend, price in each label1.3.1, 2.1.1, 3.2.2
ReviewNo review stepTable with named Change links3.3.4
PaymentUnexplained component, silent error, pale placeholderNamed region, announced error, focus on Pay, a second way to pay1.4.3, 4.1.3
Bank checkFocus stayed behind the bank's frameFrame inside a modal dialog that takes and returns focus2.1.1, 2.4.3
TimeSilent ten minute holdWarning two minutes before, one press to extend2.2.1
Confirmation"Thanks!" and a bare titleOrder number in title and heading, what happens next2.4.2

The following Saturday, Tom tries again. He changes the quantity to two and hears the new total. He checks out as a guest. The lookup still does not know his flat, so he types the address in. He chooses standard delivery with the arrow keys, checks the order on the review page, pays, confirms the code from his bank in a dialog that speaks to him, and hears "Order BB10482 confirmed, Brightwater Books." The whole route works by keyboard and by voice. Ruth has a sale she would never have known she was losing.

Notice also what Ade did not do. He did not install anything that promised to make the checkout accessible. Every change was in the shop's own HTML, its own settings and its own tests, which is the whole argument of this series, and the reason an overlay cannot do this job for you.

Conclusion

The checkout is where accessibility stops being a principle and becomes revenue. Every step is a gate, and a customer who cannot pass one leaves without telling you. Most of the gates are fixed with ordinary HTML: guest checkout, labelled fields, one radio group, a status region, a dialog, a page title that says what happened. The parts you do not own, the provider's payment component and the bank's check, still sit inside your shop, so you choose providers who document their accessibility, configure them without breaking it, frame them properly, and always keep a second way to pay.

Then you prove it, on every release: a keyboard pass, a screen reader pass, and an automated end to end test that fails where a customer would fail. If you run a shop, start with one test this week. Put the mouse away and try to buy something from yourself.

If you want to know whether every customer can actually pay you, book the free workshop: we walk your checkout together by keyboard and screen reader, step by step, and the first two weeks of work are on me.

Sources

  1. 1.

    Freeney Williams and Click Away Surveys (Rick Williams and Steve Brownlow), Press release: Click Away Pound 2019, UK retailers lose out on £17 billion by ignoring needs of disabled shoppers online, 6 February 2020, https://www.clickawaypound.com/downloads/CAP2019PR2.pdf

  2. 2.

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

  3. 3.

    NV Access, About NVDA, nvaccess.org, https://www.nvaccess.org/about-nvda/

  4. 4.

    W3C, Web Content Accessibility Guidelines (WCAG) 2.1, W3C Recommendation, 5 June 2018, updated 6 May 2025 (success criteria 1.3.1, 1.3.5, 1.4.3, 2.1.1, 2.2.1, 2.4.2, 2.4.3, 3.2.2, 3.3.1, 3.3.3, 3.3.4, 4.1.2 and 4.1.3), https://www.w3.org/TR/WCAG21/

  5. 5.

    W3C Web Accessibility Initiative, Understanding Success Criterion 3.3.4: Error Prevention (Legal, Financial, Data), https://www.w3.org/WAI/WCAG21/Understanding/error-prevention-legal-financial-data.html

  6. 6.

    UK Parliament, Equality Act 2010, section 29, Provision of services, etc., legislation.gov.uk, https://www.legislation.gov.uk/ukpga/2010/15/section/29

  7. 7.

    UK Parliament, Equality Act 2010, section 20, Duty to make adjustments, legislation.gov.uk, https://www.legislation.gov.uk/ukpga/2010/15/section/20

  8. 8.

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

  9. 9.

    CEN, CENELEC and ETSI, EN 301 549 V3.2.1 (2021 03), Harmonised European Standard, Accessibility requirements for ICT products and services, ETSI, 2021, https://www.etsi.org/deliver/etsi_en/301500_301599/301549/03.02.01_60/en_301549v030201p.pdf

  10. 10.

    Baymard Institute, Cart and checkout abandonment rate statistics, baymard.com, updated 28 October 2026, https://baymard.com/lists/cart-abandonment-rate

  11. 11.

    W3C, Web Content Accessibility Guidelines (WCAG) 2.2, W3C Recommendation, 5 October 2023 (success criteria 3.3.7 Redundant Entry and 3.3.8 Accessible Authentication (Minimum)), https://www.w3.org/TR/WCAG22/

  12. 12.

    W3C Web Accessibility Initiative, Understanding Success Criterion 3.3.8: Accessible Authentication (Minimum), https://www.w3.org/WAI/WCAG22/Understanding/accessible-authentication-minimum.html

  13. 13.

    W3C Web Accessibility Initiative, Understanding Success Criterion 4.1.3: Status Messages, https://www.w3.org/WAI/WCAG21/Understanding/status-messages.html

  14. 14.
  15. 15.
  16. 16.

    Government Digital Service, GOV.UK Design System, Addresses pattern, https://design-system.service.gov.uk/patterns/addresses/

  17. 17.

    W3C Web Accessibility Initiative, Understanding Success Criterion 3.3.7: Redundant Entry, https://www.w3.org/WAI/WCAG22/Understanding/redundant-entry.html

  18. 18.

    W3C Web Accessibility Initiative, Forms tutorial: Grouping controls, https://www.w3.org/WAI/tutorials/forms/grouping/

  19. 19.

    W3C Web Accessibility Initiative, Understanding Success Criterion 3.2.2: On Input, https://www.w3.org/WAI/WCAG21/Understanding/on-input.html

  20. 20.

    Adyen, Accessibility of Web Drop in and Components, Adyen Docs, with the Accessibility Conformance Report prepared by Level Access for version 6.40.2, https://docs.adyen.com/online-payments/accessibility

  21. 21.

    WooCommerce, Stripe for WooCommerce Accessibility Conformance Report, woocommerce.com, 21 July 2026, https://woocommerce.com/accessibility/stripe-for-woocommerce-accessibility-conformance-report/

  22. 22.

    Stripe, Authenticate with 3D Secure, Stripe Documentation (the issuer requirement, display options for the authentication interface, and the minimum authentication timeout in the mobile SDKs), https://docs.stripe.com/payments/3d-secure/authentication-flow

  23. 23.
  24. 24.

    W3C Web Accessibility Initiative, Understanding Success Criterion 2.2.1: Timing Adjustable, https://www.w3.org/WAI/WCAG21/Understanding/timing-adjustable.html

  25. 25.

    Government Digital Service, GOV.UK Design System, Error summary component, https://design-system.service.gov.uk/components/error-summary/

  26. 26.

    Government Digital Service, GOV.UK Design System, Confirmation pages pattern, https://design-system.service.gov.uk/patterns/confirmation-pages/

  27. 27.

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