What should I check before sending a PDF?

Rotation is a flag, not the pixels

Scanned and photographed documents arrive sideways constantly, and the way PDF handles that is the source of a specific and confusing failure.

Each page in a PDF carries a rotation entry: a value of 0, 90, 180 or 270 that tells a viewer how to turn the page before displaying it. The content itself is untouched. This is efficient and it means the same file can display correctly while the underlying page is stored in its original orientation.

The confusion arrives when something reads the page without honouring the flag. Text extraction, some printing paths, and a number of older systems take the raw page and produce a sideways result from a file that looked upright everywhere you checked it. The reverse also happens: a viewer that lets you rotate the view without saving gives you a document that appears fixed on screen and is unchanged in the file.

The practical rule is to rotate and save, then reopen the saved file and confirm. If a recipient reports a sideways document that looks correct to you, the flag is the first thing to check, and rotating in a tool that writes the change to the file resolves it.

Page order deserves the same treatment. Documents assembled from several sources, or scanned with a duplex feeder that produces all fronts followed by all backs in reverse, arrive in an order that no amount of careful reading will fix later. Reordering is cheap to do and expensive to discover after the document has been signed.

An unflattened form is still a form

This is the most consequential item on the list and the least understood, because a filled form looks finished on screen.

PDF form fields are interactive objects that sit on top of the page. What you see when a form is filled in is the field displaying its current value; the value is stored separately from the page content, and it remains editable. Sending a completed form without flattening it means the recipient can click into any field and change what it says, and depending on the viewer, can do so without any indication the document was altered.

Flattening converts each field into ordinary page content. The appearance is preserved exactly and the interactivity is discarded, so the text becomes part of the page in the same way a printed line is. After flattening, changing a value requires editing the document rather than typing into it.

There is a second reason to flatten that has nothing to do with tampering. Form fields render differently across viewers, because the field appearance is generated by the viewer rather than fixed in the file. A form that looks correctly aligned in one application can render with clipped text or a different font in another, and printing sometimes omits field contents entirely under certain settings. Flattening removes that variability by fixing what the page contains.

The trade-off is that flattening is one-way. If the recipient needs to fill in further fields, or if the document is a template, flatten a copy and keep the original. And note that flattening a signature field discards its cryptographic properties along with its interactivity — a digitally signed document should not be flattened, because doing so destroys exactly the thing the signature was for.

What a watermark does and does not do

Watermarks are useful and they are routinely expected to do a job they cannot do, so it is worth being precise.

A watermark is page content — text or an image drawn over or under the existing content. It communicates status: draft, confidential, the name of the recipient, a date. Applied under the content it sits behind text and stays legible; applied over it with transparency it survives being covered by an opaque element in the original. Neither placement changes what it fundamentally is.

What it does not do is prevent anything. A watermark is not encryption and not a restriction. It can be removed by anyone with a PDF editor, and content can be extracted from underneath it. Treating a watermark as protection for confidential material is a category error; if the content must not be read by the wrong person, the answer is not sending it, or sending it encrypted.

Where watermarks earn their place is deterrence and attribution. A per-recipient watermark makes a leaked copy traceable, which changes behaviour even though it prevents nothing. A DRAFT mark stops a working document being mistaken for a final one, which is the single most common document accident in any organisation.

Page numbers belong in the same part of the process and have a more mundane justification. A document assembled from several files has no numbering, or worse, has several conflicting sequences from its source documents. Anything intended to be discussed, referenced or printed needs one authoritative sequence, and legal and academic submissions frequently require it. Add numbering after the page order is final, because numbering added before a reorder is worse than no numbering at all.

The order matters, and metadata goes last

These operations interact, and doing them in the wrong sequence means redoing work or shipping a file that does not carry the changes you made.

Start with structure: remove pages that should not be there, fix the order, and correct rotation. Everything downstream depends on the page set being final, and every one of these steps invalidates numbering.

Then add content that references the structure: page numbers first, then any watermark. Numbering before the order is settled produces a document whose printed numbers disagree with its actual sequence, which is the specific failure that makes a paginated document useless for reference.

Then flatten form fields, if there are any. Flattening before adding a watermark is fine; flattening before the content is final means the fields are no longer available if something needs correcting.

Strip metadata last, and this is the step whose ordering is least obvious. Every edit writes metadata. Loading and saving a PDF in most libraries updates the producer field to name that library and the modification date to the current time, so a file scrubbed at the start of the process arrives at the end carrying a fresh set of identifying fields recording exactly what you did to it. Cleaning first and editing after is not a partial improvement; it is no improvement, because the fields you removed are simply rewritten.

The metadata worth removing is more than the author name. A PDF carries title, subject, keywords, creator and producer, creation and modification timestamps, and often an XMP packet duplicating and extending all of it. Documents converted from word processors frequently retain the original filename and the account name of whoever created it, neither of which the sender expects to be sharing.

One last check that costs nothing: open the finished file, scroll every page, and confirm it is what you think it is. Assembly errors are common, silent, and visible in seconds to the person receiving the document.

Questions