Why your rotated PDF keeps reverting

View rotation versus page rotation

Most PDF viewers have a rotate button that turns the page on your screen only. It is a display preference, like zoom. Nothing in the file changes, so closing the document forgets it and anyone else opening the file sees the original orientation.

Every page in a PDF also carries a /Rotate value stored inside the file — 0, 90, 180 or 270 degrees. That one travels. Changing it and saving is what people mean by rotating a PDF "for real", and it is a different button, often in a different menu, and sometimes behind a paywall.

Why scans are the usual culprit

Scanners and phone scanning apps write the orientation they detected into that flag. They detect it from the text, and they get it wrong on anything unusual: a landscape table, a form with sideways column headers, a page that is mostly a photograph, a document in a script the detector was not trained on.

This is also why one page in a scan can be sideways while the rest are fine. The flag is per page, not per document — which is a nuisance when a tool only offers "rotate all".

Rotation is additive, and this trips people up

The rotation you apply is added to whatever the page already had. A page already at 90°, rotated by another 90°, lands at 180 — not at 90.

In practice that means the fix for an over-rotated page is not "undo" but the complementary angle. Turned something 90° too far? Apply 270° and it returns to where it started. The four values wrap around, so 90 + 270 = 360 = 0.

Does rotating hurt quality?

No, and this is the one piece of good news in the whole topic. Rotation changes a number in the page dictionary. The text, the vectors and the embedded images are untouched — nothing is re-rendered, re-compressed or resampled. Rotate a page four times and you have the original file back, byte for byte in all the ways that matter. Compression, by contrast, genuinely does discard data; rotation does not.

Rotate a few pages, not all of them

For a scan where only some pages are wrong, target them directly rather than rotating everything and fixing the rest afterwards. Page ranges accept the notation you would expect — 2, 4-6, 9 — and it is worth reading the page numbers off the document itself rather than the viewer's counter, which sometimes starts at a cover page you were not counting.

A useful order of operations for a messy scan: rotate first, then remove blank pages, then compress. Doing it the other way around means recompressing pages you were about to throw away.