The short version
A VPAT is the report that states, criterion by criterion, how accessible your product is. People price the document and miss the point. The document is cheap. Re-auditing every release is the expensive part, and that is the cost an AI agent actually moves.
- ADA is the law, WCAG is the standard, Section 508 and EN 301 549 adopt WCAG, and the VPAT reports against all of them.
- The price scales with how many screens, how many standards, and how much real testing sits behind the verdicts.
- An agent does the per-screen grind, so re-auditing a release stops being a special project.
People ask what a VPAT costs the way they ask what a house survey costs, expecting a single number. The real answer is that the report itself is the cheap part. What costs you is everything that has to happen before anyone can write it, and the fact that you owe it all again the next time you ship.
A VPAT, short for Voluntary Product Accessibility Template, is the document that states how accessible your product is, criterion by criterion. Enterprise buyers and government agencies ask for one before they will sign a contract. So the question is real, and the straight answer has more to do with how often you change your product than with the going rate for a document.
What a VPAT actually is
The words around accessibility get tangled, so it helps to lay them out in order.
The ADA is the law in the United States, and it contains no technical spec. When a court asks whether a website met the ADA, it looks at WCAG, the Web Content Accessibility Guidelines, which is the standard with testable criteria, fifty-five of them at the AA level most teams aim for. Section 508 governs what United States federal agencies are allowed to buy, and Europe’s EN 301 549 does a similar job across the Atlantic. Both point back at the same WCAG criteria rather than inventing their own.
A VPAT, once filled in, is called an ACR, an Accessibility Conformance Report. It is where you write down how your product does against all of that. Each criterion gets one of four verdicts: Supports, Partially Supports, Does Not Support, or Not Applicable. One audit, done right, fills in every row for WCAG, Section 508, and EN 301 549 at the same time. That single report is what a buyer is really asking for.
Where the cost really comes from
Now the maths. A report is priced by scope, and scope has three dials.
The first is how many screens. Accessibility gets tested screen by screen, so a product with a handful of pages and one with hundreds are not in the same world. The second is how many standards. A WCAG-only report is the cheap edition. Add the Section 508 and EN 301 549 sections and the price climbs, because there is more to fill in. The third dial, the one that actually decides the bill, is how much real testing sits behind the verdicts. A report backed by a proper screen-reader pass on every screen costs far more than one where someone ran a scanner and waved the rest through. It should. It is worth more.
Then comes the part nobody quotes you. The report is a snapshot of one moment. The next release that moves the interface around can break things the report called fine, and now the snapshot is wrong. A VPAT is only as current as your last audit, which means a product that ships often owes the audit often. That recurring bill, not the one-off report, is where the money actually goes.
Related reading
How I ran a real accessibility audit is the work that produces these rows. Why overlays do not work is what happens when you try to buy your way past it.
How AI changes the maths
This is where an agent earns its keep, and it is not by writing the report. It is by collapsing the per-screen grind that makes re-auditing so painful.
The expensive line was always the human hours, walking every screen with a scanner, a keyboard, and a screen reader, recording what each control does. An agent can do most of that pass unattended. It runs the scanner, drives a real screen reader, measures contrast in both themes, and produces a structured list of what is wrong and where. The report then falls out of that list almost for free, because filling in fifty-five rows is the easy part once you are holding the evidence.
So the cost curve bends. The first audit is still work. But the second one, after a release, stops being a special project you budget and schedule, and turns into something closer to a test you re-run. That is the gap between auditing once and being able to afford to audit every time, and only the second version actually keeps a product accessible.
Generating the report
The last step is mechanical, and worth showing because it takes the mystery out of the document. The audit produces a structured list of problems. A small generator turns that list into the ACR, row by row, in the standard VPAT format. A few rows from a real one read like this:
| WCAG criterion | Conformance | Remarks |
|---|---|---|
| 1.1.1 Non-text Content | Supports | Images and icons carry text alternatives |
| 1.4.3 Contrast (Minimum) | Partially Supports | Most text passes; some dark-mode button labels measured below 4.5 to 1, fix in progress |
| 2.1.1 Keyboard | Partially Supports | Core flows operable; a few custom widgets not yet reachable by keyboard |
| 4.1.3 Status Messages | Partially Supports | Most updates announce; per-toggle save status not yet exposed to a screen reader |
Notice the Partially Supports verdicts and the plain remarks beside them. That is what a truthful report looks like mid-repair. Some criteria pass outright, some are part-way, each with a note on exactly what is left. A buyer can read that and know what they are getting. A wall of Supports with no remarks is the thing to distrust.
The rule you cannot skip
There is one way to make an AI-assisted VPAT worthless, and it is the same trap the overlay vendors fell into. If the report claims a screen-reader pass that never happened, it is lying, and a lie in a conformance report is a worse defect than the bugs it papers over. The agent has to record which assistive tech it actually ran on each screen, and where it only inferred behaviour from the accessibility tree instead of listening to a real screen reader. Those are not the same thing, and the remarks have to say which one you got.
This is also why a person still signs the report. An agent can do the testing and draft every row. A human reads it against the evidence and puts their name on it, because a VPAT is a claim you are making to a customer, and a claim needs someone accountable for it. I wrote up the audit that produces these rows, and why the overlay shortcut fails for the same reason a VPAT that lies does. The technology got faster. The bar for telling the truth did not move.
