Accessibility

How conformant it is

GovComms.ai is partially conformant with WCAG 2.2 Level AA. Partially conformant means some parts do not fully conform, and in this case it also means some parts have not been checked by a method capable of finding the problem.

We are not claiming full conformance, and the reason is in “What has not been tested” below rather than buried in a footnote.

What was tested, and when

A self-assessment on 20 September 2026, using axe-core 4.10.2 against the WCAG 2.0, 2.1 and 2.2 rulesets, run against live pages at 1440×900, with each route loaded as a role that can actually reach it. A separate pass at 320px width checked reflow.

24 routes were tested. 22 returned no detectable issues on the first run. That includes every page added since the previous audit on 12 September, and every page that audit fixed, which is the more useful half: those fixes have held.

This was carried out by the development team on its own product. It is a self-assessment, not an independent audit.

What was found, and fixed

Four problems. All four are fixed and the fixes are live.

• A public page linked to a route that does not serve. Our transparency page told anyone affected by a draft that the complaints process was at a web address that redirects to the home page. That is fixed, and it is the finding we would most want a reader to know was possible.
• Invalid table roles on the document list (WCAG 1.3.1, Level A). Column headers were marked up as a table that was not there, so screen readers discarded them and announced nothing. The sort state now sits on each button’s own name, where it is read out on the control you are about to press.
• Two labels on one file upload control (WCAG 1.3.1 and 3.3.2). Browsers disagree about which to announce, so what you heard depended on your screen reader.
• The draft page scrolled sideways at 320px (WCAG 1.4.10, Level AA). Content 520px wide in a 320px window, which forces horizontal scrolling to read a line of text.

What has not been tested

This is the section that qualifies everything above. None of the following has been checked, and none of it is closed.

• No screen reader has been used to test this product. There has been no VoiceOver or NVDA pass. Automated testing finds missing names; it cannot tell you whether what is announced makes sense in the order it arrives, or whether the new sort wording reads well to somebody actually listening to it.
• Focus management in the corpus analysis dialog has not been checked: whether focus moves into the dialog, stays there, and returns to the control that opened it.
• Live regions have not been checked. We do not know whether “Drafting…” is announced. If it is not, someone using a screen reader has no way to tell whether the system is working.
• The complaints page could not be tested, because it does not serve yet. See “Reporting a problem”.
• Target size and text resize were not re-measured in this pass. The September audit found them conformant and no control size has changed since, which is a reason to expect they still pass rather than evidence that they do.

One more limit worth stating plainly: automated testing is estimated to find around a third of accessibility problems. “22 of 24 clean” means no issues our tools can detect. It does not mean no issues.

Reporting a problem

If you cannot use part of this system, we want to know, and the honest position is that we do not yet have a dedicated route for telling us.

For now, raise it with the public body whose communications you received. They are accountable for what they publish and can escalate it to us. A direct accessibility and complaints route is being set up, and this page will name it, with a response time, once it is open.

We are not giving a response time before that route exists, because a published commitment nobody is on the other end of is worse than none.

Keeping this current

This statement is written from an audit dated 20 September 2026 and is only as current as that audit. Three of the four problems above were introduced by work done in the days before it, one of them four days earlier, so a statement that is not re-checked after changes stops being true quickly.

It will be revised whenever the audit is re-run. If the date above is old, treat the claims here as old too.