How we test
How we test a practice's website
This page sets out the standard we test against, the software that runs the checks, what each check does and what it can't find. You can pass it to whoever runs your website.
The standard and the rule
We test against WCAG 2.1 Level AA, the standard the HHS Section 504 rule names for websites and mobile apps (45 CFR 84.84). It has 50 success criteria, and your report gives a result for each one. HHS's fact sheet on the rule names the same standard. So does the American Dental Association's Digital Accessibility Guide for dental practices, which tells practices to follow WCAG 2.1 Levels A and AA.
The dates are May 11, 2027 for practices with 15 or more employees and May 10, 2028 for practices with fewer (91 FR 25496). Whether the rule covers your practice is a question for your attorney. Nothing here is legal advice. All the official documents
How an audit runs
We follow the steps of the W3C's evaluation method for websites (WCAG-EM): set the scope, look through the site, choose the pages, check them, then write the report.
- We choose your main pages: the home page, your contact and appointment pages, and the others patients use most. The report lists each page tested and the date.
- We test the live pages and a saved copy of them. Our software blocks the common tracking scripts, such as Google Analytics and Facebook's pixel, and it never sends your forms.
- The checks below run on every page.
- Every result is checked against the page and its pictures before it goes in the report. False alarms come out, and problems with one cause become one finding.
- You get the report within 5 business days of your order.
The checks
Two rule engines read the code of each page: axe-core 4.11.0 from Deque Systems and IBM's Equal Access checker 4.0.34. They test what can be decided from the code: text contrast, names for links, buttons, images and frames, form labels, ARIA, headings and landmarks. They can't tell whether a text alternative says the right thing, whether the reading order makes sense, or whether a control works when someone uses it.
Testing software then drives a real browser, Microsoft Edge, through each page:
- Keyboard only. A walk through the page with the Tab key in a 1280 by 900 window. At every stop the software takes a picture with focus and one without. If the two match, a keyboard user can't see where they are.
- On arrival. A second Tab walk that starts while the page is still loading, the way a keyboard user arrives.
- 200% zoom. Each page laid out as a 1280 by 900 window zoomed to 200%. It looks for content that's cut off or scrolls sideways, text that's clipped or runs over other text, and focus hidden behind a header or banner.
- Phone width. Each page at 320 pixels wide, the width of a small phone, checked for sideways scrolling and content cut off at the edge.
- Text spacing. The wider line, letter, word and paragraph spacing from WCAG 1.4.12, counting text that gets cut off.
- Moving content. Sliders, videos and animations that move by themselves, and whether each has a way to pause it.
- Forms and images. Where each field's label comes from, its autofill hint, image text alternatives, frame titles, and what a screen reader is given for each part of the page.
Then one person on our team goes through your home page and your appointment or contact form with the NVDA screen reader. What they hear goes into the report alongside the software's results.
What the software covers, criterion by criterion
These are the criteria that come up most on practice websites. "Software" means the checks above decide it. "Partly" means our checks cover part of it, and the line says what's left. "Not covered" means our checks don't reach it, and the report says so where it applies.
| WCAG 2.1 criterion | How it's checked |
|---|---|
| 1.1.1 Non-text Content (Level A) | Partly The software finds images, icons and buttons with no text alternative. Whether a text alternative says the right thing needs a judgment. |
| 1.2.2 Captions (Level A) | Not covered Someone has to watch each video. The report says when a page has one. |
| 1.3.1 Info and Relationships (Level A) | Partly The rule engines check labels, lists, tables and headings in the code. Whether the code matches what the page shows needs a look. |
| 1.3.5 Identify Input Purpose (Level AA) | Partly The software lists name, phone, email, address and birth-date fields that have no autofill hint. Whether each field asks about the patient themself needs a look. |
| 1.4.3 Contrast (Minimum) (Level AA) | Partly The software measures text on plain backgrounds against 4.5 to 1, or 3 to 1 for large text. Text over a photo or a gradient gets a look. |
| 1.4.4 Resize Text (Level AA) | Software The 200% zoom check, plus pages that stop people zooming in on a phone. |
| 1.4.10 Reflow (Level AA) | Software The 320-pixel phone-width check. |
| 1.4.11 Non-text Contrast (Level AA) | Partly The software measures form field outlines. Icons and focus rings get a look against their backgrounds. |
| 1.4.12 Text Spacing (Level AA) | Partly The software counts text that gets cut off. Text that slides over other content shows only in the pictures, which are looked at one by one. |
| 2.1.1 Keyboard (Level A) | Partly The keyboard walk shows what the Tab key can reach. Menus, sliders and other controls also have to work with Enter, Space and Esc, and those are tried one by one. |
| 2.1.2 No Keyboard Trap (Level A) | Software The keyboard walk records any place the Tab key gets stuck. |
| 2.2.2 Pause, Stop, Hide (Level A) | Software Finds things that move by themselves and checks each has a way to pause it. |
| 2.3.1 Three Flashes or Below Threshold (Level A) | Not covered Our software can't read video or animation frame by frame. The report says when a page has video or animation. |
| 2.4.1 Bypass Blocks (Level A) | Software A "skip to content" link, or headings and a marked menu. |
| 2.4.4 Link Purpose (In Context) (Level A) | Partly The software finds links with no name. Whether each link's words say where it goes needs a judgment. |
| 2.4.7 Focus Visible (Level AA) | Software A picture at every keyboard stop, with and without focus. |
| 3.1.1 Language of Page (Level A) | Software Each page has to say what language it's in. |
| 3.3.1 Error Identification and 3.3.3 Error Suggestion | Not covered We don't send your forms, so we don't see their error messages. |
| 4.1.2 Name, Role, Value (Level A) | Partly The rule engines find buttons, links, fields and frames with no name. Custom controls such as menus and sliders are tried with the keyboard. |
| Screen readers and voice control | Partly One person on our team goes through your home page and your appointment or contact form with the NVDA screen reader. Other screen readers and voice control aren't covered. |
What software can't find
Automated testing finds only part of the problems. The W3C, which publishes WCAG, says evaluation tools "can not determine accessibility, they can only assist" (Selecting Web Accessibility Evaluation Tools). In Deque's study of about 300,000 issues from first-time audits, its automated tests found 57% of them. Our browser checks catch some problems a scan misses, and the report says which check found each one.
No audit can promise that your site meets the rule. Your report says which pages we tested, on what date, how we tested them and what we found.
How each problem is written
- What's wrong, in plain words for you.
- Where it is: the page, the part of the page and the code that causes it, so your web person can go straight to it.
- How it was found: by a rule engine, a browser check, the NVDA screen reader or reading the page code.
- The WCAG criterion and its level, and how much it affects patients.
- Who controls the fix: your own site, or a vendor such as an online booking tool.
- The fix for your platform, such as WordPress, Squarespace or Wix, with the code to add or change.
- Pictures of the spot where they help, such as the keyboard stops.
See one problem the way your web person gets it
Rechecks
For 90 days after your report, we recheck the same pages free, as often as you ask. We run the same checks on the same pages and send an updated report within 5 business days, with each problem marked as no longer found or still found. If a fix we gave you turns out to be wrong, we correct it and recheck it free, any time.
What we see on health-care websites
Since November 6, 2026, our automated checks have run on the home pages of 6,793 health-care websites in 48 states: dental and medical practices, chiropractors, therapists, behavioral health, eye care and home health. On 5,809 of them (86%), axe-core found at least one WCAG 2.1 A or AA failure on the home page alone.
| What the check found | WCAG | Home pages |
|---|---|---|
| Text too faint to read against its background | 1.4.3 | 66% |
| Links with no name a screen reader can read out | 2.4.4, 4.1.2 | 49% |
| Embedded frames, such as maps and videos, with no title | 4.1.2 | 19% |
| Images with no text alternative | 1.1.1 | 18% |
| Pages that stop people from zooming in on a phone | 1.4.4 | 12% |
Counted on November 7, 2026 from axe-core 4.11.0 results on one page per website, so these are only the problems an automated check of the home page can find. We don't name any practice.
We test this website too
Its latest run, on November 7, 2026, covered 23 pages and page states in light and dark mode. axe-core 4.14.0 found no WCAG 2.1 A or AA failures, no page scrolled sideways at 320 pixels wide, and on the 9 pages walked with the keyboard every Tab stop showed a visible focus outline. The check runs before we publish a change. It doesn't include a person testing with a screen reader. How we checked this site
Who we are
Northshelf LLC is a small company in Clinton Township, Michigan. Our Michigan LLC ID is 900305608; you can look it up on the State of Michigan's business registry. Call us at (586) 217-0032, 8 AM to 6 PM Eastern, seven days a week, or email hello@northshelfaudits.com. Relay calls through 711 are welcome.