Extracting a Few Pages from a Statement Without Installing Anything

Extracting and deleting are not two routes to the same place

A PDF is a numbered collection of objects — page descriptions, fonts, images, streams of drawing instructions — with a table recording where each lives and which are currently referenced. The list of pages is itself just another object, a tree pointing at the page objects in order. Removing a page from that tree is a small edit: the page stops being displayed and stops being counted. Whether the object it pointed at is also erased from the file is a separate decision, and many editors do not make it, particularly when they save by appending changes to the end of the file rather than rewriting it.

Extraction has no such ambiguity, because it works in the opposite direction. It creates an empty document and copies in only the pages you asked for, along with the resources those pages need. Nothing is left behind because nothing else was ever brought across. That is the whole reason to prefer it: the guarantee comes from the construction, not from trusting a tool to have cleaned up after itself.

The black rectangle is the most expensive mistake here

It is worth being blunt about this because it has embarrassed governments, law firms and newspapers, repeatedly and publicly. A PDF page is drawn in layers, and an annotation drawn last sits on top of what came before. Placing a filled black rectangle over an account number changes what the page looks like and changes nothing about what it contains. The text underneath is still an object in the file. It can be selected with the cursor, copied out, extracted by any text-extraction tool, and read by a search engine that indexes the document.

The reliable way to remove something from a page you are keeping is to stop the text from existing. Turn the page into an image, paint over the region in that image so the pixels themselves change, and build a new PDF from the result. You lose selectable text on that page, which is the point: there is no longer any text to select. It is a real cost — the page is no longer searchable and no longer readable by assistive technology — so apply it to the pages that need it and not to the whole document.

Page numbers on the page are not page numbers in the file

This catches almost everyone once. Extraction counts physical pages from one; the number printed on the page counts whatever the document's designer decided. A statement with an unnumbered cover sheet puts printed page 1 at physical page 2, so asking for pages 3 to 5 gets you printed pages 2 to 4 — plausible enough that you may not notice, and wrong enough that the recipient will. Annual reports with roman-numbered front matter shift by more, and a document assembled from several sources can restart numbering partway through.

The cure is to read the physical position, not the printed label. Scroll to the page you want and note where it sits in the viewer's counter, which counts physically. Then extract, open the result, and confirm you got the content you meant rather than the numbers you asked for. On a document whose pages all look alike — twelve months of the same statement layout — that confirmation is the only thing standing between you and sending February when you were asked for March.

What travels with an extracted page, and what does not

The page's own content comes across intact — text, fonts, images, vector drawing. Things that belong to the document rather than to the page usually do not. Bookmarks pointing at pages you did not take have nowhere to go. Internal links, a table of contents, cross-references to other sections all break or disappear. Form fields may survive as fields or may be flattened depending on the tool. If any of that matters for what you are sending, check it on the extracted file rather than assuming.

One thing that does travel, and should not be forgotten, is metadata. Author, producing application, creation and modification dates and sometimes a full editing history live at the document level and are frequently copied to the new file. On a document you are sending outside your organisation, that can disclose more than the pages do — the name of the person who prepared it, the internal template it came from, the fact that it was modified an hour ago.

Send the pages, or send the whole thing?

There is a real tension here, and the honest answer depends on who is asking. A landlord, an employer or a lender asking for proof of one thing has no claim on the rest of your financial life, and sending only the relevant pages is both reasonable and, under data-minimisation principles, the better practice. A tax authority or an auditor asking for a period will usually want the whole period, and extracting the parts you consider relevant looks like selection — the appearance of choosing your evidence can cost you more than the disclosure would have.

A practical middle path covers most cases: send the complete document for the period requested, with anything genuinely irrelevant and sensitive removed at the pixel level rather than by omitting pages. That keeps the sequence intact and verifiable — page numbers run without gaps, balances carry forward — while still not handing over what nobody asked for. Whichever you choose, say what you did in the covering message. "Pages 3 to 5 of the March statement" invites no suspicion; a silent gap between page 2 and page 6 invites plenty.

Check the result, not the operation

Every failure in this article is invisible in the tool and visible in the output. The extracted file has the wrong pages; the page you redacted still yields text when you drag the cursor across it; the metadata still names your colleague. All three take seconds to check on the file you are about to send, and none of them can be checked on the file you started from. Open the result, count the pages, try to select the thing you hid, and only then attach it.

Frequently asked questions

Can the recipient tell that I extracted only part of the document?

Usually yes, and that is fine. Printed page numbers, a header saying "page 3 of 12", a running balance that starts partway through — all of these make it obvious, and none of them is a problem when you have told them what you sent. Trying to disguise the extraction is where it becomes one: a partial statement presented as a complete one is a much worse thing than a partial statement described as partial.

Does extracting reduce the file size proportionally?

Roughly, but not exactly, and it can surprise you in both directions. Fonts and images shared across the whole document have to be carried along if any extracted page uses them, so three pages out of twelve may come to more than a quarter of the original. Conversely, if the pages you dropped held all the large images, three pages may come to almost nothing. Judge by the resulting file, not by the fraction of pages.

How do I check whether a redaction actually worked?

Open the finished file, drag the cursor across the blacked-out area, and copy. Then paste into any text box. If anything appears, the information is still in the document and the black shape is decoration. Do the same with the whole page selected — sometimes the hidden text is reachable that way even when a direct drag looks empty. If the page has been turned into an image, nothing will be selectable anywhere on it, which is both the confirmation and the cost.

The file is password-protected. Can I still extract pages?

Not until the protection is removed, and that requires the password you were given. The tool cannot read the page structure of an encrypted document, so it cannot copy pages out of it. Bank statements arrive this way routinely, usually with a password derived from a date of birth or a customer number. Remove the protection, extract, and then decide whether the extract needs protecting again — for something being uploaded to a portal over an encrypted connection, usually not.

Is splitting a large document a good way to get under an upload limit?

Often the best one, and better than degrading the file, because most forms cap each attachment rather than the whole submission. Two halves of 6 MB upload where one 12 MB file will not, and neither half has lost any quality. It is particularly the right answer when the document must stay at full resolution because a recognition system will read it. Name the parts so the order is obvious — "statement 1 of 2" — and say in the message that it is split.