Will a screen reader actually read your PDF?

No. The file is read into your browser's memory and checked by code running in this page. It is never sent to GotAlt or to any other server, and nothing is kept once you close the tab. You can verify that yourself: open your browser's developer tools, switch to the Network tab and run a check. You will see this site's own files being loaded (the PDF reader, plus its character maps and standard-font files when a document needs them) but no request that carries your PDF or its name. Those map and font files are named after what the document uses, so a Japanese document asks for a Japanese character map: GotAlt can see that kind of detail in its request logs, never the file. Like every page on this site, this one also loads a web font from Google Fonts; that request is about the font and has nothing to do with your file.

Up to 100 MB, because your browser has to hold the whole file in memory. Within that, page-level checks (text, tags, figures, tables and form fields) cover the first 200 pages. Document-level checks (title, language, bookmarks and security settings) always cover the whole file. If a document is longer than 200 pages the results say it was only partly checked.

A tagged PDF contains a hidden structure that labels each piece of content as a heading, a paragraph, a list item, a table cell, a figure and so on, in reading order. Screen readers rely on it. An untagged PDF is closer to a picture of a page that happens to contain text: the reader has to guess the order. Getting tags right usually means exporting again from Word or InDesign with tagging switched on, which is far less work than repairing the tags afterwards.

The tool decides by whether it can extract any text. Text that has been converted to outlines (some design and print exports do this) looks like a picture to it. To test a flagged file yourself, open it in a PDF viewer and try to select a word with the mouse. If you cannot, a screen reader will not be able to read it either. The check can also be fooled the other way: text in a font with no Unicode map counts as text, even though what comes out may be nonsense. Copying a sentence and pasting it into a text editor shows which case you have.

No. It means the checks this tool can run found nothing wrong. Alt text can be present and useless, the reading order can be wrong and colour contrast is not checked at all. Treat a clean result as a good start, then run PAC (Windows only) and veraPDF and listen to the file with a screen reader.

PAC is a free checker built around PDF/UA and WCAG that runs on Windows only, and veraPDF is an open-source validator that covers PDF/UA. Adobe Acrobat Pro also has a built-in accessibility check, but it is a paid product. We built this tool for the quick, private first look, and we would rather you use those for the full check than pretend this replaces them.

The Department of Justice's rule for state and local governments treats a PDF on a public entity's website as web content. Its exception for older documents covers only documents that were posted before the compliance date, and not ones people still use to apply for, get access to or take part in a service. Read the DOJ's own fact sheet, and our guide to the Title II deadline, which tracks the dates. This is not legal advice.

That number is our own rule of thumb, not something WCAG sets. A short document is easy to move around in; in a long one, bookmarks work as a table of contents that keyboard and screen reader users can jump through. If a shorter document is hard to find your way around, add bookmarks anyway. The W3C technique for PDF bookmarks explains how they help.