Compare PDFs

Compare PDFs is a text comparison, not a visual one. Each document is opened by an in-page extractor, the positioned text runs on every page are reassembled into visual lines, and the two line lists are matched with a longest-common-subsequence diff — the classic algorithm behind source-control diffs. What that catches: inserted, deleted and edited wording, and pages that exist in one file only. What it cannot see: a changed font, a different colour, a resized or replaced image, a shifted table, a new logo, or anything else that alters the picture without altering the characters. Two files that read identically but look completely different come back with no differences, and for this tool that is the correct answer. A visual check still means putting the two renderings side by side yourself.

Line-level diffing is the reason: changing one word reflows the rest of the paragraph, and every reflowed line is a different line. The comparison works on lines rebuilt from the PDF's text runs, so if inserting a word pushes three words onto the next line, every following line of that paragraph differs from its old version and shows up as a red block followed by a green block. This is inherent to comparing formatted output rather than the source document. Such a diff reads much better as before-and-after paragraphs than as word edits. Two other things bend the result: extraction order depends on how the producer wrote the text stream, so reordered runs can look like moves, and each page is capped at 4,000 lines to keep the comparison responsive.

No — the two files can be different lengths, and pages that exist on one side only are labelled Page added or Page removed. The comparison is positional: page one is matched against page one, page two against page two, and so on until the longer document runs out, at which point the leftover pages are reported as added or removed outright. Presence is tracked separately from content, so a deleted blank page is still reported as a deletion instead of quietly pairing off against another blank page. The limitation worth knowing is what positional pairing costs you: insert a new page three into the updated file and every page after it lines up against its neighbour, producing a large but meaningless diff. When that happens, compare the documents in sections instead.

Scanned PDFs can only be compared once they contain a real text layer. A scan is a picture of a page, so there are no characters in the file for the extractor to find; both sides come back with no text and the tool correctly reports that nothing changed on any page. The fix is to run each file through PDFChamp's OCR PDF tool first, which adds an invisible recognised-text layer, and then compare the two OCR'd copies. Expect some noise in that diff: recognition errors differ between two scans of the same page, so a handful of lines will show as changed purely because one pass read a character differently. Encrypted files behave similarly — when the content streams cannot be decoded, no text is extracted and the diff comes back empty rather than wrong.

The Download diff (.txt) button saves the entire comparison as a plain-text patch named pdf-diff.txt. It opens with a header naming both files and the totals — pages compared, lines added, lines removed, lines unchanged — followed by one section per page, marked with the page number and, where relevant, a note that the page was added in the updated PDF or removed from the original. Inside each section, changed lines carry a plus or minus prefix while unchanged lines are indented by a single space, so the file reads like any diff you would meet in a code review. That makes it easy to paste into an issue tracker, attach to a message to a counterparty, or keep as a record of what changed between two revisions of an agreement.

Comparing two documents here costs nothing and sends neither of them anywhere. Both files are read from disk by the browser, their text is extracted inside the tab, and the diff is computed in memory — there is no comparison server, no queue and no account. That is the whole point for the documents people actually compare: draft settlements, employment agreements, tender responses and board papers, which are precisely the files you should not be posting to a stranger's backend. There is no usage limit and no paid tier that unlocks longer documents; the only ceilings are your device's memory and the 4,000-line-per-page cap that keeps very long pages responsive. The .txt patch is generated locally and carries no branding of any kind.