How to choose samples that represent the audited service.
Why sampling matters
Auditing every page of a non-trivial service is rarely possible. The goal of sampling is to choose a set small enough to audit thoroughly but representative enough that the result generalizes.
Sampling dimensions
Pick samples that vary along these axes:
- Page type — homepage, navigation index, content page, search results, form, error state, confirmation, login, account settings, media-heavy page.
- Layout pattern — different templates and component compositions.
- Authentication state — public vs. logged-in.
- Content state — empty state, populated state, error state.
- Dynamic behavior — pages with significant JavaScript-driven interaction.
- Service Criticality – pages within the critical (essential) logic of your service. (product detail, support, checkout, order tracking, etc.)
A sample set that covers all distinct page types and layouts is more useful than one that picks ten pages of the same type.
Sample size guidance
For a small website/service:
- Minimum: one sample per distinct page type.
- Common: 10 to 15 samples covering the main flows.
For medium service (under ~50 unique templates):
- Minimum: one sample per distinct page type + critical paths pages.
- Common: 16 to 25 samples covering the main flows and key edge cases.
For larger services:
- Identify product-line clusters, sample within each.
- Augment with journey-based sampling. See journey mapping.
Sampling and journeys
Page samples and journey maps can overlap. A journey passes through samples that are evaluated in your audits. This is desirable: standalone evaluation catches per-page failures, journey evaluation catches cross-page failures and overall accessibility coverage.
Documenting sampling decisions
Record the reasoning in the audit's scope notes:
- What types are represented.
- What is intentionally excluded.
- Why a particular flow was prioritized.
This makes the audit defensible when challenged. You can take these notes at the audit level, or at the sample level to justify each sample selection.