Skip to content
Ada EditorStart writing

Research

WCAG 2.1 AA in meeting documents: what software can decide, and what needs a person

Of the 22 WCAG 2.1 Level A and AA criteria that come up in every agenda, packet or set of minutes, software can decide none on its own. It can catch some failures of eight, ten need a person, and the file format usually takes care of the other four. That’s our own sorting, criterion by criterion, below.

Published
Last reviewed

The short answer

Title II’s rule will hold most documents that state and local governments post online to WCAG 2.1 Level A and AA, from April 2027 or April 2028: 50 success criteria. In a meeting document most of them depend on what’s in it, such as video or a fillable form. The 22 that come up every time are about text, headings, lists, tables, links, images and colour.

W3C puts it plainly: “Tools cannot check all accessibility aspects automatically. Human judgement is required.” For these 22, software is most useful as a list of where to look. Ada Editor checks eight of them; the table names each check.

The criteria in every meeting document

CriterionWho can decide itIn Ada Editor
1.1.1 Non-text Content (A)Partly software. Software can tell an image has no alt text. Whether the alt text says what the image means takes a person.Images without alternative text; Alternative text that may not describe the image; Images that may need a long description
1.3.1 Info and Relationships (A)Partly software. Software can tell a document has no structure at all: a PDF with no tags fails whenever it has headings, lists or tables. Whether a bold line is really a heading, or a table really holds data, takes a person.Heading levels that skip; Empty headings; Lines that look like headings but aren’t; No headings at all; Fill-in blanks; Tables without header cells; Tables that may only lay text out; Tables with merged cells
1.3.2 Meaningful Sequence (A)A person. Software can show the reading order. Whether it makes sense around columns, text boxes and footnotes takes a person.Export PDF tags the content in the document’s own order
1.3.3 Sensory Characteristics (A)A person. An instruction such as “see the map on the right” or “the shaded parcel” takes a person to notice.Not checked
1.3.4 Orientation (AA)The file format. A Word file or PDF can’t lock the screen to portrait or landscape; the viewer turns it.Not checked
1.4.1 Use of Color (A)A person. Software can flag words like “in red”. A chart or map that uses colour alone takes a person to spot.Instructions that rely on colour
1.4.3 Contrast (Minimum) (AA)Partly software. Software measures text on a plain background. Text over a photo or a gradient takes a person.Text contrast
1.4.4 Resize Text (AA)The file format. Readers zoom in their viewer. W3C’s guidance for documents says a viewer that zooms to 200% meets it, unless the content doesn’t work with that zoom.Not checked
1.4.5 Images of Text (AA)A person. Software can point to pages with no text in them, such as scans. Whether a picture is a picture of text takes a person.Not checked
1.4.10 Reflow (AA)A person. Someone narrows a web page to 320 pixels and reads it. For a PDF, W3C’s guidance for documents warns that where a format’s viewers don’t reflow, meeting it “may not be possible”.Not checked
1.4.12 Text Spacing (AA)A person. It applies where readers can change the spacing, as on a web page; W3C’s guidance for documents leaves out formats where they can’t, as in most PDFs. A person widens the spacing and looks for text that’s cut off.Not checked
2.1.1 Keyboard (A)A person. Every link, and every field in a form, has to work from the keyboard. Someone tabs through them with the keyboard alone.Not checked
2.4.2 Page Titled (A)Partly software. Software can tell a file has no title, or only the default an authoring tool gave it, such as “Document1”. Whether a real title identifies the document takes a person.Titles that don’t say what the document is. Export PDF writes the document’s title (or “Untitled document”) into the file and asks viewers to show it
2.4.3 Focus Order (A)A person. Whether moving through the links with the Tab key follows a sensible order takes a person.Not checked
2.4.4 Link Purpose (In Context) (A)Partly software. Software can find a link with no words at all, and point to “click here” and bare web addresses. Whether a link’s words, with their sentence, say where it goes takes a person.Links like “click here”; Links that are a bare web address; The same link text going to different places
2.4.6 Headings and Labels (AA)A person. Whether each heading describes its section takes a person.Not checked
2.4.7 Focus Visible (AA)The file format. The viewer or browser shows which link has focus, unless a web page’s styles hide it.Not checked
2.5.3 Label in Name (A)Partly software. Software can compare the words a link or field shows with its name in code. Labels in pictures take a person.Not checked
3.1.1 Language of Page (A)Partly software. Software can tell whether a language is set. Whether it’s the right one, it can only ask: a guess from the text isn’t certain, least of all in a bilingual agenda.The document’s language
3.1.2 Language of Parts (AA)A person. Software can point to passages that look like another language. Whether they are, or are names and borrowed words, takes a person.Passages in another language
4.1.1 Parsing (A)The file format. The current WCAG 2.1 says it “should be considered as always satisfied” for HTML and XML, which covers a web page and a Word file, and W3C’s guidance for documents says the same of a PDF. The Title II rule cites the 2018 printing, which predates that note.Not checked
4.1.2 Name, Role, Value (A)Partly software. Software can tell a link or form field has no name or role. Whether its state and value are announced as they change takes a person.Not checked

Every Ada Editor check is explained in Help, with how to fix it. What its exports keep is in Help’s section on exporting.

The ones that depend on what’s in it

The other 28 apply when a document has one of these. Ada Editor checks none of them.

A recording, video or audio

  • 1.2.1 Audio-only and Video-only (Prerecorded) (A): a person. A transcript or description exists or it doesn’t, but whether it matches the recording takes a person.
  • 1.2.2 Captions (Prerecorded) (A): a person. Software can tell whether a video has a caption track, but not captions burned into the picture, or whether they’re accurate.
  • 1.2.3 Audio Description or Media Alternative (Prerecorded) (A): a person. Whether the description covers what’s shown takes a person.
  • 1.2.4 Captions (Live) (AA): a person. Live captions are judged as they happen.
  • 1.2.5 Audio Description (Prerecorded) (AA): a person. Whether the description covers what’s shown takes a person.
  • 1.4.2 Audio Control (A): a person. Software can find sound that plays on its own. Whether it can be paused or stopped is checked by trying.
  • 2.3.1 Three Flashes or Below Threshold (A): software. Flash-analysis tools check video frame by frame.

A fillable form

  • 1.3.5 Identify Input Purpose (AA): partly software. Only for formats that can mark what a field asks for, such as a web form; W3C’s guidance puts document formats without that out of scope. There, software can check the fields that ask for a name, address or phone number say so in code.
  • 2.1.2 No Keyboard Trap (A): a person. Someone tabs through the form and checks they can leave every field.
  • 3.2.1 On Focus (A): a person. Checked by moving through the fields and watching what happens.
  • 3.2.2 On Input (A): a person. Checked by filling in the fields and watching what happens.
  • 3.3.1 Error Identification (A): a person. Checked by making mistakes on purpose.
  • 3.3.2 Labels or Instructions (A): partly software. Software can find fields with no label. Whether the instructions are enough takes a person.
  • 3.3.3 Error Suggestion (AA): a person. Whether an error message says how to fix it takes a person.
  • 3.3.4 Error Prevention (Legal, Financial, Data) (AA): a person. Whether a submission can be checked, corrected or undone is checked by trying.

Moving, timed or interactive content

  • 1.4.13 Content on Hover or Focus (AA): a person. Pop-ups that appear on hover are checked by using them.
  • 2.1.4 Character Key Shortcuts (A): a person. Only content that adds its own single-key shortcuts, checked by trying them.
  • 2.2.1 Timing Adjustable (A): a person. Whether a time limit can be turned off or extended is checked by trying.
  • 2.2.2 Pause, Stop, Hide (A): a person. Whether moving or updating content can be paused is checked by trying.
  • 2.5.1 Pointer Gestures (A): a person. Swipes, pinches and drawn paths need a single tap or click as well, checked by trying.
  • 2.5.2 Pointer Cancellation (A): a person. Checked by pressing and sliding off a control.
  • 2.5.4 Motion Actuation (A): a person. Only content worked by shaking or tilting, checked by trying.
  • 4.1.3 Status Messages (AA): a person. Whether a screen reader hears a message that appears without moving focus is checked by listening.

Charts, maps or diagrams

  • 1.4.11 Non-text Contrast (AA): a person. Which lines and shapes are needed to understand a graphic takes a person. Their contrast can then be measured.

A set of documents

W3C’s guidance for documents reads these four as applying to a set of documents, not to one on its own: see where documents differ.

  • 2.4.1 Bypass Blocks (A): a person. Whether readers can skip what repeats from document to document takes a person.
  • 2.4.5 Multiple Ways (AA): a person. Whether there’s more than one way to find a document in the set takes a person.
  • 3.2.3 Consistent Navigation (AA): a person. Whether navigation that repeats stays in the same order takes a person.
  • 3.2.4 Consistent Identification (AA): a person. Whether things that do the same job are named the same way takes a person.

Where documents differ

WCAG was written for web pages. W3C’s WCAG2ICT note explains how each criterion reads for documents and software outside a web page. It is guidance, not a requirement: in its own words it “does not set requirements”. For four criteria, Bypass Blocks (2.4.1), Multiple Ways (2.4.5), Consistent Navigation (3.2.3) and Consistent Identification (3.2.4), it reads “web pages” as documents in a “set of non-web documents”, so they concern a set rather than one document on its own.

The laws differ too. Section 508, for federal agencies, uses WCAG 2.0 and says non-web documents “shall not be required to conform” to those same four. Title II’s rule, 28 CFR 35.200, names “Level A and Level AA success criteria and conformance requirements specified in WCAG 2.1” and lists no criteria as excepted. Who it covers, and from when, is in our guide to Title II for meeting documents.

What a checker misses

When we walked a clerk’s path through Ada Editor, with an invented agenda shaped like the ones in our agenda platform research, its nine section titles were bold capitals, not headings, and Ada Editor flagged none of them. Headings are 1.3.1, and they’re how screen reader users skim to their item (our record of the walkthrough).

We added a check that asks about short lines in bold or heading-sized type: lines that look like headings but aren’t. It asks rather than fails, because a bold line can also be a name. A line set only in capitals, or only underlined, still isn’t counted. A checker that finds nothing has only shown it found nothing.

What a checker can fix

Very little, in our own test. When we first tested our rules, in September 2026, we ran 13 of them over 28 public documents from GitHub: READMEs, codes of conduct and guides, not meeting documents. Of 670 findings, 32 (4.8%) came with a fix software could apply, and 28 of those were the same one, setting the document’s language. The other 638 needed what only the author knows: what the image shows, where the link goes, how to say it more plainly (our record of that test).

So most findings in Ada Editor lead with Go to text. Apply fix appears only where software can make the change, and for some, such as a line that looks like a heading, you still decide whether to use it.

Beyond AA

Ada Editor also checks two criteria that Title II doesn’t require, as advice rather than failures: no top-level heading (2.4.10 Section Headings, AAA), reading level (3.1.5 Reading Level, AAA) and long sentences (3.1.5 Reading Level, AAA).

How we sorted them

Each criterion went into one of four groups, by what software can decide without a person:

  • Software: it can pass or fail the criterion on its own.
  • Partly software: it can catch some failures for certain; a person decides the rest.
  • A person: software can at most point to where to look.
  • The file format: a Word file, a tagged PDF or a web page, opened in an ordinary viewer, usually meets it.

The sorting is ours, read against W3C’s Understanding WCAG 2.1 and WCAG2ICT, and each row says why. “Every meeting document” is our generalisation from the packets in our research. We used WCAG 2.1 because it’s what Title II names; WCAG 2.2 adds criteria not covered here. For tagged PDFs, our guide to what “ADA compliant PDF” can mean gives the PDF Association’s own count of which PDF/UA failures software can find.

Corrections and limits

Another reviewer could sort some rows differently, especially the partly ones. This page isn’t legal advice, and no checker, ours included, can confirm a document meets WCAG.

Think a row is wrong? Report a correction to this page (public, on GitHub).

None yet.

Sources

Read on October 12, 2026.