Remove metadata from a PDF file, using exiftool and qpdf. Note that embedded objects may still contain metadata.

README.md

Anonymising PDFs

PDF metadata

Metadata in PDF files can be stored in at least two places:

• the Info Dictionary, a limited set of key/value pairs
• XMP packets, which contain RDF statements expressed as XML

PDF files

A PDF file contains a) objects and b) pointers to those objects.

When information is added to a PDF file, it is appended to the end of the file and a pointer is added.

When information is removed from a PDF file, the pointer is removed, but the actual data may not be removed.

To remove previously-deleted data, the PDF file must be rebuilt.

pdftk

pdftk can be used to update the Info Dictionary of a PDF file. See pdftk-unset-info-dictionary-values.php below for an example. As noted in the pdftk documentation, though, pdftk does not alter XMP metadata.

exiftool

exiftool can be used to read/write XMP metadata from/to PDF files.

• exiftool -all:all => read all the tags.
• exiftool -all:all= => remove all the tags.

exiftool -all:all also removes the pointer to the Info Dictionary, but does not completely remove it.

qpdf

qpdf can be used to linearize PDF files (qpdf --linearize $FILE), which optimises them for fast web loading and removes any orphan data.

Embedded objects.

After running qpdf, there may be new XMP metadata, as it extracts metadata from any embedded objects. To read the XMP tags of embedded objects, use exiftool -extractEmbedded -all:all $FILE.

pdftk-unset-info-dictionary-values.php

<?php

$file = 'example.pdf';

// get the current metadata
$command = sprintf('pdftk %s dump_data', escapeshellarg($file));
$output = array(); $return = null; exec($command, $output, $return);

//print_r($output);

if ($return) {
    throw new Exception('There was an error reading metadata from the PDF file');
}

// set any metadata values to null
foreach ($output as $index => $line) {
    if (strpos($line, 'InfoValue:') === 0) {
        $output[$index] = 'InfoValue:';
    }
}

// write the updated metadata to a file
$metadataFile = tempnam(sys_get_temp_dir(), 'pdf-meta-');
file_put_contents($metadataFile, implode("\n", $output));

// create a new PDF using the updated metadata
$tmpFile = tempnam(sys_get_temp_dir(), 'pdf-tmp-');
$command = sprintf('pdftk %s update_info %s output %s',
    escapeshellarg($file), escapeshellarg($metadataFile), escapeshellarg($tmpFile));
$output = array(); $return = null; exec($command, $output, $return);

if ($return) {
    throw new Exception('There was an error writing metadata to the PDF file');
}

// clean up the temporary files
rename($tmpFile, $file);
unlink($metadataFile);

remove-pdf-metadata.sh

#!/bin/bash

FILE=example.pdf

# read tags from the original PDF
#exiftool -all:all $FILE

# remove tags (XMP + metadata) from the PDF
exiftool -all:all= $FILE

# linearize the file to remove orphan data
qpdf --linearize $FILE

# read XMP from the modified PDF
#exiftool -all:all $FILE

# read all strings from the modified PDF
#cat $FILE | strings > $FILE.txt

# read XMP from embedded objects in the modified PDF
#exiftool -extractEmbedded -all:all $FILE

exiftool (...) does not completely remove it
qpdf (...) removes any orphan data
So [these tools] cannot remove all metadata in a PDF document? Or that only using them together can do it?

The combination seems to work (but privacy is hard). My understanding is:

exiftool -all:all= puts a metaphorical sticky note on the end that says "by the way, the Author field is actually blank"

but the data is still there in the original place in the bytes of the PDF, and

exiftool -pdf-update:all= is another sticky note with "nevermind, disregard that previous sticky note".

qpdf --linearize looks at the giant pile of sticky notes and says "this is a mess, let's rewrite the entire PDF as it should have been".
From the point of view of a standards compliant PDF parser, the resulting output is identical, but now the underlying data has been changed in line with the stickynotes.

So qpdf doesn't know what changes you want, and exiftool doesn't know how to rewrite a clean PDF. But together they can make the change and get a clean PDF at the end. (I think.)