First, the thing that explains most rejections on its own: the two sides are running different checkers. Acrobat's built-in accessibility check and the PDF/UA-1 standard (ISO 14289-1) are not the same bar, and they are not close. veraPDF, the open-source reference validator for PDF/UA-1, checks 106 rules. Acrobat checks a shorter, friendlier list.
- The document has no text in it at all
- It was printed to PDF instead of saved as PDF
- The PDF/UA flag was written before the file earned it
- Alt text was invented for things that are not pictures
- The fonts are named but not carried
So a vendor runs Acrobat, sees green, and ships. The client's reviewer runs veraPDF or PAC, sees failures, and rejects. Both are telling the truth. Here is how to find out which specific failure you have.
The document has no text in it at all
If the PDF is a photograph of paper, scanned on an office copier, there are zero characters inside it. There is nothing to tag, because tagging attaches structure to text, and there is no text. No remediation tool can fix this. Not mine, not anyone's. It needs OCR first to create a text layer, and machine OCR makes mistakes that a person has to catch.
This is far more common than people expect. A 119-page municipal comprehensive plan I measured was 9.3 MB and contained zero readable characters. A city's monthly financial reports, 159 pages across seven files, were all copier scans with nothing inside them.
The free check on this site reports the exact character count per page, which also catches the common case of a typed document with a few scanned pages inserted into it.
It was printed to PDF instead of saved as PDF
This one is nearly deterministic, and it is the most useful thing on this page because the fix is free. Every PDF records which program produced it. That field predicts whether the file has accessibility tags with startling reliability.
When you use File > Print and choose a PDF printer, the print driver receives only a picture of the page. Headings, lists, table structure, alt text and reading order are all discarded before the PDF is written. There is nothing left to tag. When you use File > Save As > PDF (or Export), Word hands that structure over intact.
| Producer recorded in the file | What made it | Tagged? |
|---|---|---|
Acrobat PDFMaker for Word | The Acrobat ribbon in Word | Yes |
Microsoft® Word | Word's own Save As PDF | Yes |
PScript5.dll | File > Print to a PDF printer | Never |
KMBT_223, SKM_C360i, RICOH IM | An office copier | Never, and no text either |
The PDF/UA flag was written before the file earned it
There is a metadata identifier a file can carry that declares "this document conforms to PDF/UA-1." Some tools write it as a matter of course at the end of a remediation run.
Writing that flag does not make a document conform. It makes a document claim to conform. If the file still has untagged content or missing alt text, the flag turns a quiet problem into a loud one: now the validator reports a false conformance claim, and a reviewer who sees that stops trusting the whole file.
This is the single most common reason a file passes on the vendor's side and is rejected on the client's. The vendor's tool wrote the flag and reported success. The client's validator read the flag, checked the claim, and found it false.
pdfuaid identifier, or a conformance-claim failure, the flag was stamped early.
The fix is to remove the flag, fix the real failures, and only then write it back.
Alt text was invented for things that are not pictures
A validator reporting "figure missing alternate text" does not mean the document contains a photograph. It means something is tagged as a figure. Those are different claims, and the difference decides whether the fix is right or wrong.
A real example. The Pennsylvania State Board of Nursing's Request Form for Testing Accommodations reported fourteen figures with no alternate text. Fourteen missing descriptions sounds like a document full of undescribed images. It was not. All fourteen were horizontal rules: the fill-in lines and section dividers, between 2.9 and 4.1 points tall.
Writing fourteen captions for fourteen lines would have satisfied the checker and made the document worse: a screen reader user would hear fourteen descriptions of decoration. The correct fix is to mark them as artifacts, which is what decoration is. Same green checkmark, opposite outcome for the person actually using it.
The fonts are named but not carried
A PDF can reference Arial without containing Arial. Open it on a machine that has Arial and it looks right. Open it anywhere else and the viewer substitutes something else, and character positions shift. Under PDF/UA this is a failure, because the file does not actually specify what it renders.
This one is genuinely fixable but has a real constraint: embedding a font requires that the font's own licence permit embedding. That permission is recorded inside the font file. It is checkable, and it is not always granted. This is one of the rules that honestly cannot always be cleared from the PDF alone, without the original source file or a licence.
There is a related trap worth knowing. A subset font with no character map extracts as raw
glyph numbers rather than letters. I have measured a file whose text came out as
/0 /1 /2 /3. The document looked perfect on screen and was unreadable to every
machine. It was, as it happens, the attendee list for an accessibility procurement conference.
What a finished file should come with
Whoever does this work for you (me, another vendor, or your own team), the deliverable should include the validator report next to the file, not just an assurance. veraPDF's output is machine-readable and names every rule it checked. If your client rejected the last version, that report is what you forward to them.
Nine government documents I have remediated are published with their before-and-after measurements, including the Board of Nursing form described above, which went from nine failed rules to zero of 106: see the measured files.
What this page is not
- I hold no accessibility certification, and I do not author VPATs.
- I do not test with screen readers or other assistive technology, and I will not claim to. Conformance to PDF/UA-1 is measurable; usability by a real person is a broader question that needs people who use those tools daily.
- Passing a validator is not the same as meeting a legal obligation. Neither the ADA nor Section 508 certifies individual documents. Whether your obligation is met is a judgment for your counsel, not for a checker and not for me.
- Nothing here is legal advice.
If you have a file that keeps coming back and none of the five above explains it, send it over and I will tell you what the validator says, at no charge and with no obligation. That is a measurement, not a sales call. If the answer is that your current vendor is doing it right, I will say so.
If you already know which of the five you have and you just want the file fixed, a single document is listed at a fixed price here, paid by card into escrow and released when you accept the file, with the veraPDF report included. For an inventory rather than one file, the page rates are published. To find out which of the five you are looking at, run the free check.