WCAG 2.2 Demo SuiteStep 3 — Select a representative sample
This is the step that decides whether your audit is worth anything, and the one most often done carelessly. Two auditors with the same skill and the same standard will reach opposite conclusions about the same site if they sample differently.
The failure mode this step exists to prevent
Audit the home page and two content pages, find nine issues, write a report. The checkout — where the real barriers are, and where failure costs the user money — is never opened. The report is not wrong about what it looked at. It is worthless as a statement about the site.
3a. The structured sample
Chosen deliberately. It must include:
- Every common page you found in Step 2a — home, contact, sitemap, accessibility statement, login, search results, 404.
- At least one page of every page type from Step 2c.
- Every essential task, end to end. A checkout is not one page; sample every step of it, because the failures hide in the transitions.
- Every page with unusual interaction from Step 2e — the drag-and-drop, the map, the media player, the CAPTCHA.
- The states, not just the routes. A form has an empty state, a filled state, an error state and a success state. They are different pages as far as accessibility is concerned.
- At least one page using each web technology from Step 2d.
3b. The randomly selected sample
Then add a random sample — WCAG-EM suggests roughly 10% of the size of the structured sample, chosen without looking at the pages first.
This is not padding. Its purpose is to catch what your structured sample systematically misses: pages built by a different team, old pages predating the design system, content authored by someone who did not know the rules. If a random page fails in a way no structured page did, you have learned that your structured sample was not representative — which is itself a finding.
Select it properly: number the site's URLs from a crawl or a sitemap and pick with a random number generator. "Some pages I clicked on" is not a random sample.
3c. Include complete processes
WCAG has a specific rule here, and it catches people out. If a page is part of a process — a checkout, a multi-step application — then every page in that process must conform, or none of them can be claimed as conforming. You cannot sample step 2 and step 4 and skip step 3.
3d. Compare with the client's expectation
Show the sample to the client before auditing. Two things come out of this: they tell you about pages you did not know existed, and they cannot later say you audited the wrong things.
Worked example: Riverline Books
The sample site has eight pages. A structured sample for it looks like this:
| Page | Why it is in the sample |
|---|---|
| Home | Common page; carousel; rich media |
| Catalogue | Page type: listing with filters and sorting |
| Product | Page type: detail; images; custom widgets |
| Basket | Essential task, step 1; data table; quantity controls |
| Checkout | Essential task, step 2; form-heavy; error states |
| Login | Common page; authentication; CAPTCHA |
| Contact | Common page; form; modal dialog |
| Help | Common page; long-form content; help mechanism placement |
Eight of eight, because the site is small. On a real site of 4,000 pages the same reasoning yields perhaps 15 to 25 pages — and the discipline is identical.
A test of your sampling
The defects in the sample site are seeded so that a correct structured sample finds them and a lazy one does not. Several of the most serious — a keyboard trap, a drag-only control, a checkout with no error prevention — are on pages you only reach by working through an essential task. If you sample only the pages linked from the home page, you will miss them, and your report will say the site is in better shape than it is.