Getting text back out of a PDF means rebuilding what it threw away.
A PDF stores where each glyph sits, not which paragraph it belongs to. That is why converting one to DOCX or ODT is reconstruction rather than translation: the converter has to work out from coordinates alone which runs of text form a line, which lines form a paragraph, and which blocks are a table. Columns, footnotes and scanned pages are where it is hardest, so check those first in the result.
PDF was designed in 1993 so a document looks identical on every screen and printer. It does that by fixing every letter, line and picture to an exact position, with its fonts packed inside. Peel the layers of a typical page apart to see what's in one.
0.86 0.15 0.15 rg72 700 468 4 re fBT /F1 24 Tf 72 660 Td(Quarterly report) Tj ETbuildPDF to Word or PowerPoint turns these commands back into editable boxes and text.
A PDF doesn't know about paragraphs. Each line is placed separately at fixed coordinates, so nothing reflows when a word gets longer. Type a replacement for the highlighted word and watch the line.
Underneath the layers, a PDF is five kinds of numbered objects plus an index of where each one starts, counted in bytes from the beginning of the file. This one is generated from your settings, and it’s a real file that opens in any viewer. Change the text and every byte offset in the cross-reference table shifts to match.
%PDF-1.7%âãÏÓ
1 0 obj<< /Type /Catalog /Pages 2 0 R >>endobj
2 0 obj<< /Type /Pages /Kids [3 0 R] /Count 1 >>endobj
3 0 obj<< /Type /Page /Parent 2 0 R /MediaBox [0 0 595 842] /Resources << /Font << /F1 5 0 R >> >> /Contents 4 0 R >>endobj
4 0 obj<< /Length 43 >>streamBT /F1 24 Tf 72 746 Td (Hello, world) Tj ETendstreamendobj
5 0 obj<< /Type /Font /Subtype /Type1 /BaseFont /Helvetica >>endobj
xref0 60000000000 65535 f 0000000015 00000 n 0000000064 00000 n 0000000121 00000 n 0000000256 00000 n 0000000349 00000 n
trailer<< /Size 6 /Root 1 0 R >>startxref422%%EOF
A PDF editor doesn’t rewrite the file when you save. It appends the changed objects, a new cross-reference section and a trailer pointing back to the previous one. That is what lets a signature stay valid after someone fills in a form, and why deleted text can often be recovered. Apply the edits below in order.
An incremental save adds new objects to the end of the file and leaves the old ones where they were. This page follows the xref chain and separates the file into its revisions, listing what each one added, changed or deleted. It extracts the page text at each step, so you can see what the document said before it was edited. Signatures are checked for which bytes they cover, whether the covered bytes match the digest inside the signature, and what was changed afterwards. Every earlier revision can be downloaded as it was. Everything happens on your device. Both samples are real PDFs built in your browser.
They're all ordinary PDFs that follow extra rules. Pick who's asking to see which rules apply and what they forbid.
A PDF that opens perfectly can still be rejected, because several industries require a named subset of the format rather than the format itself. The differences are not cosmetic: each subset forbids things the general format allows, so a file can be a valid PDF and an invalid PDF/A at the same time, and the only way to know is to check rather than assume.
PDF/A is the archival subset, and its central rule is that the file must be self-contained. Every font has to be embedded rather than referenced, because a font available today may not be installed on the machine that opens the file in twenty years. Encryption is not allowed, and neither is anything whose rendering depends on the outside world. That is what makes it the format records offices and courts ask for.
PDF/X exists for print and fixes the variables a press cares about. Colours are pinned to a defined space so what comes off the plate matches what was approved on screen, trim and bleed boxes are declared explicitly, and transparency is handled predictably. A print shop asking for PDF/X is not being awkward; they are removing the ambiguity that produces an expensive reprint.
PDF/UA is the accessibility subset, and it is about structure rather than appearance. The content is tagged with what each part is, so a screen reader can announce a heading as a heading and read a table by row and column instead of reciting coordinates. Reading order is declared rather than inferred. Public sector work frequently mandates it, and a visually identical untagged file will fail.
Converting into or out of a PDF is where these requirements bite, because a conversion rewrites the file and can quietly drop what made it compliant. If a document has to meet one of these standards, check the result after converting rather than trusting that the label survived, and keep the compliant original: a PDF/A file re-saved by an ordinary tool is usually no longer PDF/A.
Identifiers and structure from ISO 32000, with each version’s headline additions. Click a version to see what it introduced; the header line of any PDF says which one it claims.
Each tool works on the layer it names above.