Working with PDFs: merge, split, extract and convert

PDF is the format people reach for when a document has to look the same everywhere — and the format they curse when they need to get anything back out of it. Both reactions come from the same design decision, made back in the early 1990s: a PDF describes exactly how each page should be drawn, and almost nothing else. Once you understand that, every PDF task — merging, splitting, extracting text, converting pages to images — becomes predictable, including the cases where it can't work.

What a PDF actually contains

A PDF is not a flat picture of a document, even though it renders like one. Inside the file, each page is a list of drawing instructions: place this run of text at these coordinates in this font, draw a line from here to there, fill this rectangle, paint this embedded image inside this box. The fonts themselves are usually embedded in the file, along with any photos or logos as compressed image data. A PDF viewer replays those instructions to paint the page.

Notice what's missing from that list: paragraphs, tables, headings. A PDF doesn't know that three text runs form a sentence or that a grid of lines is a table — it only knows where each fragment sits on the page. Tools that extract text or tables from PDFs are reconstructing structure that the format threw away, which is why the results range from perfect to mangled depending on how the file was produced.

The two kinds of PDFs

Before doing anything with a PDF, it helps to know which of two families it belongs to, because they behave completely differently.

Born-digital PDFs were exported from software — Word, a browser's print dialog, an invoicing system. These contain a real text layer: actual character codes positioned on the page. You can select text in a viewer, search it, copy it out. Text extraction works because the text is genuinely there.

Scanned PDFs came from a scanner or a phone camera. Each page is one big photograph wrapped in PDF packaging. There are no characters in the file at all, just pixels that happen to look like characters to a human. The quick test: try selecting text in your viewer. If your cursor sweeps a highlight across words, there's a text layer. If you can only drag a selection box around the whole page, it's a scan.

This distinction decides what's possible. Extracting text from a scan requires OCR — optical character recognition, software that looks at the pixels and guesses the characters — which is a separate, error-prone step that no simple extraction tool performs. If a "the text extraction returned nothing" problem lands on your desk, the file being a scan is the explanation nine times out of ten.

Merging and splitting

These are the most common PDF operations for a reason. Merging is for the situations where someone insists on receiving one file: a rental or visa application assembled from a form, an ID scan and a bank statement; a contract plus its three signed annexes; a semester's worth of lecture notes bound into a single document for reading on a tablet. Splitting is the reverse pressure — you have a 40-page report and only page 12 concerns the person you're emailing, or a scanned book chapter needs to be broken back into handout-sized pieces.

Because merging and splitting only rearrange whole pages, they're lossless: the drawing instructions, fonts and images for each page are copied over untouched, so nothing is re-rendered or recompressed. Our PDF merge and split tool does both in the browser — the files never leave your machine, which matters more for PDFs than for most formats, since they tend to be contracts, medical records and other things you'd rather not upload anywhere.

Turning pages into images

Sometimes you need a page as a picture: to drop into a slide deck, post somewhere that only accepts images, or send to someone whose phone mangles PDF previews. This is rasterization — the same replaying of drawing instructions a viewer does, but saved to a pixel grid instead of a screen. PDF to PNG renders every page as a separate image and packages them into one ZIP.

The one decision that matters here is resolution. Render at roughly screen resolution and an A4 page comes out around 800 pixels wide — fine for a slide, blurry if anyone zooms in. Render at print-quality DPI and the same page is over 2,000 pixels wide with crisp small text, but each PNG may be several megabytes. Rasterizing is also a one-way door: the output is pixels, so text in the image can no longer be selected, searched or resized cleanly. Keep the original PDF; treat the images as a derived copy.

Pulling out just the text

When you want the words and nothing else — to quote a passage, feed a document into a script, or search across a pile of files — rendering images is the wrong tool. PDF to TXT reads the text layer directly and writes plain text, typically shrinking a multi-megabyte document to a few kilobytes. Expect the layout to flatten: columns can interleave, headers and page numbers land mid-stream, and hyphenated line breaks stay hyphenated. That's the structure problem from earlier — the PDF never knew about columns in the first place. For getting searchable, greppable words out of a born-digital PDF, though, it's the fastest route there is. On scans it will return little or nothing, for the reasons above.

Making PDFs from images

The reverse direction is just as common: you photographed a signed form, or you have a folder of receipts, and the recipient wants "a PDF". Converting JPG to PDF or PNG to PDF wraps each image in a page, and multiple images become multiple pages in one document — which pairs naturally with the merge tool when the pages come from different sources. Be clear about what this does and doesn't do: the result is a scanned-style PDF, pixels in a wrapper, with no text layer. It satisfies "send me one PDF" perfectly; it does not make the content searchable.

Why PDF file sizes vary so wildly

Two ten-page PDFs can differ in size by a factor of a hundred, and the reasons are usually visible once you know what's embedded. A text-only document carries little beyond its drawing instructions and font files — often well under 100 KB. Add embedded fonts for a CJK language and the floor rises, because a Korean or Japanese font can carry thousands of glyphs (subsetting, which embeds only the characters actually used, is what keeps this sane). Photos dominate everything else: a PDF stores each image with its own compression, so a document built from ten uncompressed phone photos can pass 30 MB while the same photos, JPEG-compressed at reasonable quality before conversion, produce a file a tenth the size. If a PDF you've built from images is too big to email, compressing the images first is the lever that matters.

When not to use PDF

PDF is an output format — a way to freeze a finished document. It's the wrong home for anything you'll need to edit again. A budget exported to PDF can't be recalculated; a report sent only as PDF forces the next editor to reconstruct it by copy-paste. The habit worth building is simple: keep the source file — the XLSX, the DOCX, the Markdown — and generate a fresh PDF from it whenever you need to share. The PDF is the printout, not the document. Treat it that way and every tool on this page stays a convenience instead of a rescue operation.