How we measured.
Before building CivicFormat we wanted to know what a screen reader actually meets in the documents governments post. This page says what we looked at, how we checked it, and what the numbers can’t tell you.
The wider survey
We looked at agendas and minutes ( pages) from city and county websites in states.
have no accessibility tags
have no real title
are pictures of pages
of the tagged files we checked meet the PDF/UA standard in an automated check
How it was done
- Sites. A random sample of city and county governments from the public .gov registry that CISA publishes. The sample is fixed by a seed, so it can be repeated. We looked up of them, and the ones with agenda or minutes files in the archive are the ones counted here.
- Documents. Files whose web address contains “agenda”, “minutes” or “packet”, found in Common Crawl’s public web archive (crawl ), up to per site. We read the archive’s copy, so no government’s server was visited. Common Crawl cuts files off at 1 MB, so only files it captured whole were counted.
- Checks. Whether the file has tags, a real title (not a file name) and a language, and whether it is a scan with no readable text. For the tagged files we also ran veraPDF, the open-source PDF/UA validator (a strict test: it checks the whole PDF/UA standard, not just whether the file reads in order).
- Reporting. Totals only. No government is named here or anywhere on this site.
Run on .
Our first look
We pulled 21 public PDFs from the website of a Tennessee county: monthly commission agendas, minutes and notices. Of the 21, 19 had no document title, which is the first thing a screen reader announces. All six main meeting agendas came out of the agenda software with no tags at all. One agenda was a scanned copy with no text a screen reader could read.
That is a small sample from one place. It told us where to look, not how common it is, and it’s why we looked wider.
What these numbers can’t tell you
- Tags aren’t the same as readable. A file can have tags and still read badly, and an untagged file can still be partly usable. These checks find the common starting problems; a person checking the pages is what finds the rest.
- An automated check can’t prove a document works for everyone. That is why CivicFormat has a person check every page, and why we never call a document “compliant”.
- The archive isn’t everything. It holds what a crawler happened to capture, and large files are cut off, so the sample leans toward smaller documents.
- Sites, not people. Each site counts alike. A large county and a small town weigh the same.
Questions about the method? Write to hello@civicformat.com.