What a VPAT Is, and Why Nobody Can Generate One For You

What it actually is

The Voluntary Product Accessibility Template is a form published by the Information Technology Industry Council. You go through the accessibility standard one success criterion at a time and, for each, state whether your product Supports, Partially Supports or Does Not Support it — with remarks explaining why.

The filled-in version is properly called an Accessibility Conformance Report. Almost nobody says that. When a buyer asks for your VPAT they mean the completed report.

For WCAG 2.1 Level AA that is fifty rows. Not fifty checks — fifty judgements.

Why you are suddenly being asked

It is a procurement gate, not a legal requirement. Federal agencies buying software have to satisfy Section 508, so they push the question to vendors. State agencies, universities and school districts have copied the practice. Increasingly so have private companies large enough to have a procurement function, and funders attaching conditions to grants.

The practical effect is that the document sits between you and the money. That is why it gets attention that the underlying accessibility work never did — which is a poor reason to care about accessibility and an excellent reason to finally get it done.

Why no tool can honestly write one for you

Automated testing evaluates somewhere around a third of the WCAG success criteria. That is not a limitation of any particular scanner; it is what the criteria are. A machine can tell you an image has no alt attribute. It cannot tell you whether alt="image" describes the photograph. It can find text below the contrast threshold. It cannot decide whether your page’s reading order makes sense, whether a custom control is operable by keyboard, or whether an error message actually explains the error.

So any tool offering a finished VPAT is doing one of two things. Either it is marking criteria as Supported that nothing tested — in which case you are signing a conformance claim on evidence that does not exist — or it is quietly narrowing the document to the part it can see and hoping the buyer does not notice.

The first is the dangerous one. A false ACR is worse than no ACR. It goes out with your organisation’s name on it, to a funder or a plaintiff’s lawyer, and it is a written statement that you tested something you did not.

What a tool can honestly do

It can do the boring two-thirds. Building the fifty-row table, mapping every automated failure to the criterion it belongs to, and writing the remarks for those rows is hours of clerical work that a machine does perfectly and a person does resentfully.

And it can be explicit about the rest. A criterion with no automated coverage should say Not Evaluated in the document itself, not be left blank or, worse, assumed. That tells your reviewer exactly where to spend their time, and it tells the buyer exactly what the document is worth.

What we changed

Any scan can now produce a VPAT draft covering WCAG 2.1 Level A and AA — all fifty criteria, with each automated failure mapped to the criteria it breaks and the failing element counts written into the remarks.

Nothing in it is ever marked Supports. A criterion whose automated checks passed is marked Partially Supports, with a remark saying a reviewer must still judge the rest. A criterion no automated test covers is marked Not Evaluated. On a typical page that is twenty-three of the fifty rows left openly unanswered.

It downloads as an editable document with a header stating plainly that it is a machine-generated draft requiring human completion. Delete that header when a qualified reviewer has finished the work — not before.

If you have been asked for one today

Scan the product, take the draft, and give it to whoever knows the software best. They will find the twenty-three unanswered rows are mostly questions they can answer in an afternoon — is there video, is there a time limit, does the keyboard get trapped — because they concern things a person knows and a crawler cannot see.

If the answer to several of those is genuinely unknown, that is the real finding, and it is better to learn it from your own draft than from the buyer’s reviewer.