Compare two PDFs and see what changed

In two ways at once, because either alone would mislead you. Every page of both documents is rendered and reduced to a fingerprint, which is what catches a moved image, a replaced chart or a different stamp — changes that do not alter a single word. Separately, where both pages carry a text layer, the words are compared and the exact additions and deletions are reported. A page counts as changed if either check finds something.

No, and this is the thing most comparison tools get wrong. Comparing page one to page one and page two to page two falls apart the moment a page is inserted, because every later page shifts and the tool reports the whole document as rewritten. Here the pages are matched to each other first, wherever they have moved to, so an inserted page is reported as one addition and the pages around it stay unchanged.

No. Both files are opened, rendered and compared inside your browser, and neither one leaves your device. That matters more here than for most tools: comparing two versions usually means comparing two drafts of something confidential — a contract, an offer, a set of accounts — and the whole point of doing it here is that you do not have to hand both versions to a third party to find out what changed.

No. Both documents are opened for reading only and nothing is written back. There is no output file to download, because the comparison is something you look at rather than something that replaces your documents. Your original and your revision are exactly as they were when you finish.

It is still compared, visually, and the tool tells you that is what happened rather than quietly claiming the text matched. A scanned page is a picture, so there are no words to compare — but a picture that changed still looks different, and that is detected. If you need the words compared too, run the scans through OCR PDF first to give them a text layer, then compare the results.

Usually because something moved very slightly, or the file was regenerated by different software that places glyphs a fraction differently. The tool reports how much of each page differs, so a page showing a fraction of a percent is telling you the change is tiny. What will never mark a page as changed is metadata: two files with different titles, authors or timestamps but identical pages are reported as unchanged, because the document is what you are comparing, not the file.

Up to 300 pages per file. Every page of both documents has to be rendered before anything can be compared, so the work grows with the page count and an unbounded job would exhaust a phone rather than finish. If your documents are longer, split them and compare the parts.

It is a word-level comparison, and it is honest about being one. It finds which words were added and which were removed, which is what you want for a changed figure, a reworded clause or a new paragraph. It does not understand meaning, so moving a sentence somewhere else on the page reads as a deletion and an insertion rather than a move. Whitespace is ignored, because a PDF has no lines — where a line breaks is a property of the file rather than of what it says.

Not directly — both inputs have to be PDFs. If you have a Word file, export it to PDF first and compare that. Comparing formats rather than documents would mean guessing at how the Word file would have been laid out, and any difference the guess introduced would be indistinguishable from a real edit.

Yes. Open the tool once while connected so the rendering engine is stored on your device, and after that comparison runs with no network at all. There is nothing to send and nothing to fetch — the whole comparison happens in the browser tab, which is also why it keeps working on a train.