Why deleting a page does not shrink a PDF

A PDF is not a stack of pages

It is a flat pool of numbered indirect objects. Some are numbers or strings, some are dictionaries, some are streams of compressed bytes. At the end of the file sits a table of byte offsets pointing at every object, and a trailer pointing at the table.

The document catalog points to a page tree, and each page is a leaf on that tree with its own content stream, resources and dimensions. A page's position in the document is determined by exactly one thing: its index in its parent's Kids array.

That single fact explains reordering, deleting and merging. All three are edits to an array of references.

Ten pages, delete one, measure it

Here is a ten-page document before and after removing page two, counted rather than described.

The file lost seven bytes. Seven is the length of the reference "12 0 R " that was taken out of the Kids array. The page object itself, its content stream and its resources are all still in the file.

And the content is recoverable. Pulling the deleted page's content stream straight out of the "after" file gives its drawing operators intact, including the hex-encoded string of the text that was on it. The page is invisible, not gone.

This is a privacy problem before it is a size problem

The size is a curiosity. The consequence is not: deleting a page and emailing the file can still ship the content of that page to whoever receives it.

The same applies to two other things people do for the same reason. Cropping sets a display region and leaves everything outside it in the file. Drawing a black rectangle over text leaves the text underneath, still selectable, still extractable — which is how several well-publicised redaction failures happened.

If the point is that someone must not be able to read something, none of these operations achieves it. Extract the pages you want into a new document instead, which copies only what you selected.

Incremental saving is the other reason files grow

PDFs support saving by appending. An edit can be written as a block of new objects plus a new cross-reference table pointing back at the old one, leaving the original bytes untouched at the front of the file.

So a document edited three times can contain all three previous states, and each save can make it larger. It is why a file that has been through several rounds of revision is often far bigger than its content justifies, and why old versions of a document are sometimes recoverable from it.

What actually makes a PDF smaller

Garbage collection: walk the object graph from the catalog and drop everything unreachable. This is the pass that turns a deletion into a real reduction, and the common browser-side library does not have one.

Downsampling images: on a scan or a photo document this is most of the file, and it is the only lever with a large effect. On a text-and-vector document images may be under ten per cent of the bytes, which is why "compress" does almost nothing to some files and halves others.

Subsetting fonts: dropping unused glyphs. On a typical academic paper the fonts are a bigger share of the file than the images.

Anything calling itself a compressor is mostly doing the second of those. Content streams are already Flate-compressed by every producer, so compressing them again buys almost nothing — the general-purpose entropy coding has already been spent.

Answers

Why did my PDF not get smaller after deleting pages?

Because deleting a page removes a reference to it, not the page itself. Its objects stay in the file, unreachable but present, until something garbage-collects them.

Can someone recover a page I deleted?

Often yes. The content stream is still there and can be extracted with ordinary tools. To genuinely remove pages, build a new document from the pages you are keeping.

Does cropping remove the cropped part?

No. Cropping sets the region a reader displays. Everything outside it remains in the file and can be restored in one click.

Why does compressing a text PDF do nothing?

Because compression is mostly image downsampling, and a text document has few images. Its bytes are in fonts and already-compressed content streams.