How we test a website
A person tests your website against WCAG 2.1, Level A and Level AA, all 50 success criteria. An automated scan finds approximately one third of the barriers. The other two thirds are found by hand, and every finding in the report was confirmed by a person.
Which pages
We test a sample that covers each page template and the path a patient takes to book or to contact you. A usual practice website has eight to twelve pages. A page has content on it, or a form on it. A light page has little content, or it repeats a pattern we tested already. The report names each page we tested, and the price names each page before the work starts.
The patient's task
Every audit follows one task: find the way to book or to contact the practice, and complete it. It also follows the information a patient needs in order to attend: the hours, the location, the insurance accepted, and the privacy notice. Each finding is rated by what it does to that task, not by the element on its own.
Six passes, about ten hours
- An automated scan of each page with axe-core, at desktop and phone widths. Every result is confirmed or removed by a person in the next passes.Pass 1
- The keyboard. Every control on every page, with the Tab key and Shift and Tab. Can it be reached, is the focus visible, is the order right, can focus leave it, does Enter operate it.Pass 2, 1 to 2 hours
- The screen reader. Each page read top to bottom, then by headings and by landmarks, then the task itself. What is announced is copied into the report.Pass 3, 2 to 3 hours
- The checks a machine cannot make. Zoom to 200 and 400 percent, alt text that is right, link text that says where it goes, headings that describe, motion, target size, color alone, error messages. Then the criteria the scan never touches, one by one, until all 50 have a result.Pass 4, 2 to 3 hours
- Sorting. Each finding gets a severity from the rule below, with the reason written down, and a count of instances.Pass 5, 1 hour
- Judgement and the write-up. Each finding gets its title, what fails, what it does to a patient, the correction with the markup after it, and a screenshot with the element outlined.Pass 6, 2 to 3 hours
The severity rule
- Critical. On the tested pages, a patient who uses the named assistive technology cannot complete the task at all. A form field with no label, a submit button with no name, a keyboard trap on the path to the form.
- Serious. The patient can complete the task only by guessing, backtracking or getting help. Or the information a patient needs in order to attend cannot be reached at all. A navigation link with no name, a section title that is not a heading, a notice published as a picture.
- Moderate. The patient completes the task and reaches the information, and something is harder or announced wrongly on the way. Low contrast on text that is not the task, a decorative image announced as an image.
The scanner has its own ratings. They rate the element. This rule rates the patient's task, and the report says so.
Your markup and your supplier's
A booking widget, a map or an advertisement from a vendor has its own failures. We list them, marked as the supplier's, so you can send them on. Your number is your own markup alone.
What you get
A report as a web page and as a tagged PDF: a title page, a table of contents, a summary with the overall recommendation, then each finding in a table with five rows: the finding, its severity, its location, what a screen reader announces, and the detailed remediation recommendation. After the findings: all 50 criteria with their result, the pages tested, the method, the scope limitations, and the next steps. Then a 30-minute meeting to go through it. Read the sample report to see one.
The tools
axe-core 4.12.1 in a Chromium browser through Playwright. The keyboard. The browser's accessibility tree. Browser zoom at 200 and 400 percent. NVDA on Windows for the screen reader pass. A Windows machine and a Mac, so the two families of browser are used.
What a scan cannot see
A scan cannot tell whether alt text is right, whether a heading describes its section, whether the focus order makes sense, or whether a page can be used at 400 percent zoom. A count of zero from a scan is not proof: a parked page and a page that refuses bots also scan clean. Thus every count in our reports is a minimum, and every finding was seen by a person.
After the corrections
Your developer makes the corrections. We test the same pages again at half the first price, and the re-scan report marks each finding corrected or not corrected, with the same evidence.