A sampled audit proves the pages somebody opened, and it says nothing about the pages nobody did. That is the short answer, and on its own it misleads, because the popular version of it overshoots in the other direction. You will be told that a whole-site conformance claim is not allowed. It is allowed. WCAG says so in as many words, and then attaches a condition that an ordinary sample cannot meet.
The condition is the part worth knowing, because it is the thing you can actually act on. A claim covering many pages holds where every page inside it was itself evaluated, or was produced by a process that guarantees each one passes. Two routes, and no third. So the question is never how many pages were sampled. It is whether the pages your sentence covers went through one of those two routes.
The short version
Conformance is defined only for web pages. A conformance claim may still cover one page, a series of pages such as a checkout, or many related pages such as a website. What it may not do is cover pages that were never evaluated and were not produced by a process guaranteeing they pass. A twelve-page site tested end to end can carry a site-wide sentence. A twelve-page sample of a four-hundred-page site cannot, and no larger sample fixes it.
Conformance Is Defined for a Page, and a Claim Can Still Cover a Site
Two sentences sit next to each other in the standard, and reading either one alone produces a wrong answer. The first says conformance is defined only for web pages. The second says a conformance claim may be made to cover one page, a series of pages, or multiple related web pages. People who have read only the first tell you a site can never be claimed. People who have read only the second sell you a certificate.
Both sentences are doing work. The first sets the unit of measurement, which is why what counts as a web page turns out to matter so much to a quote. Every page either satisfies the criteria at the level in question or it does not, on its own, with no partial credit and no averaging across neighbours. The second sentence then says a claim is free to name as many of those units as you like. A claim is a description of a set of pages, and the set can be big.
What links them is that the claim inherits its truth from the pages inside it rather than the other way around. Naming four hundred pages in a claim does not confer anything on those pages. It only asserts something about each of them, four hundred times over, and each assertion is separately true or false. So the size of the claim is your choice and the truth of it is not.
The Condition That Decides Whether Your Claim Holds
W3C's evaluation methodology states the condition plainly, and it is short enough to memorise. A claim covering a series of pages or a website applies when all the pages in scope have each been evaluated, or created in a process that ensures they each satisfy all the conformance requirements. That gives you exactly two ways a page earns its place inside your sentence.
- Somebody opened it and tested it. The page went through an evaluation against every conformance requirement at the level being claimed. This is the route most people picture, and it is the expensive one, because it scales linearly with the page count.
- A process guarantees it. The page came out of a pipeline whose output is known to satisfy the requirements. A locked template with no author-editable regions, a generator whose output has been evaluated, a component library where the component was tested and the page is nothing but instances of it.
The second route is the one worth thinking hard about, because it is where a large site can honestly get further than its budget suggests. It is also the one that quietly fails, because the guarantee has to hold for the whole page rather than the parts you built. A template that produces perfect markup still hands you a page whose body copy an editor wrote this morning, complete with an unlabelled image and a heading level skipped for looks. The process guarantees the frame. It does not guarantee the contents.
Why No Sample Size Closes the Gap
The methodology gives a reason for the sampling limit, and the reason is stronger than the one most people assume. It is not that samples are chosen carelessly, or that ten pages is too few and forty would be enough. It is that on any site large enough to need sampling, it is always possible that a page nobody examined carries an error. That possibility does not shrink to nothing as the sample grows, so there is no page count at which the claim becomes safe.
W3C then says something about its own method that a vendor would never volunteer. Because most uses of the methodology evaluate a sample, most uses of it do not produce a WCAG conformance claim at all. That is the standards body describing the ordinary result of its own procedure, and it should reset what you expect a normal audit to hand you. What you get is findings, evidence and a scope statement. What you do not get, unless every page was covered, is a sentence about the site.
There is a second reason the arithmetic is worse than it looks, and it catches people who thought they had counted carefully. A full page includes each variation the page presents automatically at different screen sizes, and each of those variations has to conform on its own for the page to conform. So the mobile layout of a page in your sample is not a bonus. It is part of what that one page count already promised.
One Site, Three Sentences, and Only Two of Them Hold
The abstract version of this argument is easy to nod along to and hard to use on a Monday. So here is a worked case. Take a four-hundred-page site with a shop attached, audited across twelve pages chosen to cover the templates and the checkout. Three sentences somebody might write afterwards, and what each one would actually need behind it.
| The sentence somebody wants to write | Does the sample support it | What would |
|---|---|---|
| Our website conforms to WCAG 2.2 Level AA. | No. It covers 388 pages nobody opened. | Evaluating all 400, or narrowing the sentence to the pages you did evaluate. |
| The checkout process conforms to WCAG 2.2 Level AA. | Yes, if every page in the process was tested and passed, start to finish. | Nothing further. A process is a legitimate claim unit, and this one was tested whole. |
| These twelve pages conform to WCAG 2.2 Level AA, and the templates behind them are shared with 300 more. | Yes for the first half. The second half is a true statement about templates, not a claim about those 300 pages. | Nothing, as long as the two halves stay visibly separate and the 300 are never described as conforming. |
The third row is where most honest sites end up, and it is worth defending rather than apologising for. It tells a reader something real about coverage, it tells your own team which fixes reach furthest, and it does not assert anything you would have to withdraw when somebody finds a broken page. The failure mode is not the modest sentence. It is the modest sentence quietly promoted to the first one in a sales deck.
What a Sample Does Buy, Which Is More Than a Claim
None of this makes a sampled audit a weak purchase, and it would be strange to sell one while pretending otherwise. A conformance claim is optional in the standard, and most of the reasons people commission an audit have nothing to do with making one. Here is what the twelve pages genuinely deliver.
- Template evidence. A failure found in a shared header is a failure on every page carrying that header, and fixing it once clears all of them. The sample is where you find out which failures are of that kind.
- Journey evidence. A process tested start to finish is a complete claim unit in its own right, and it is usually the part of the site carrying the revenue or the statutory duty.
- A defensible scope statement. A report naming what was tested, when, and on which combinations is evidence you can hand a customer, a procurement team or a regulator. Nobody has to take your word for the parts you did not cover.
- A price you can plan against. Sampling is what makes an audit cost hundreds rather than tens of thousands, and the cost of an audit tracks the page count almost directly.
When a Whole-Site Claim Is Genuinely Available to You
Small sites are the case everybody forgets, and if you run one this section is the whole page for you. The methodology recommends evaluating the entire product wherever that is feasible, and says the sampling procedure may simply be skipped. It names two situations: a product with a small number of views, and a product that cannot meaningfully be split into views at all.
So a fourteen-page brochure site, fully tested, can carry a sentence about the whole site that a two-hundred-page sample of a large one cannot. That is not a loophole. It is the same rule producing a different answer, because the facts are different. It is also why a small business often ends up with stronger evidence than an enterprise that spent thirty times more.
The other route is the process guarantee, and it rewards organisations that invested in structure years before anyone mentioned accessibility. If your entire catalogue is generated from one evaluated template, with no free-text region an author can break, then the catalogue is inside the guarantee. Work out where the guarantee stops before you write the sentence, because that boundary is the real edge of your claim.
Write the Scope Sentence Before You Write Anything Else
If you are going to make a claim, the standard tells you what has to be in it, and the list is short enough to check in a minute. Getting these five things down first also has a useful side effect, which is that it forces the scope decision into the open while it is still cheap to change.
- The date of the claim.
- The guidelines title, version and address, written out rather than referred to.
- The conformance level satisfied, so A, AA or AAA.
- A concise description of the pages covered, which the standard says must state whether subdomains are included.
- A list of the web content technologies relied upon.
The fourth item is the one that does the work here, and the subdomain instruction is not a footnote. It is the standard making you decide, in writing, whether the shop on a different hostname is inside your sentence or outside it. Most overclaiming we see is not a lie about a tested page. It is a claim whose fourth item was never written down, so it spread to cover things nobody meant it to.
Our own audits are samples, which is why our method lists that limit first among the things it does not cover, and why pricing says the same in nearly the same words. What comes back is a true account of the pages we opened plus a map of the templates behind them. What nobody can hand you is a certificate covering the addresses they never loaded.
Where This Page Stops
Everything above is about what the standard defines and what evidence can support. It is not legal advice, and the difference matters here more than it looks, because a conformance claim is a statement your organisation publishes and can be held to. Whether a particular sentence satisfies a particular contract, procurement rule or national regulation is a question about that document and your jurisdiction, and it is a lawyer's question rather than an auditor's.
What this page can do is make sure the conversation with that lawyer starts from an accurate description of what was tested. Get the scope sentence right first. If you want to work out how many pages a useful sample would be for your site, read how many pages an audit should include next. W3C's evaluation methodology is where the sampling rules come from.