How to Compress a PDF Without Wrecking It

An email bounces at 10 MB, an upload form refuses anything over 5, and the PDF you have is 24. The instinct is to run it through a compressor and hope. It is worth thirty seconds of diagnosis first, because the right fix depends entirely on what made the file big.

Text is never the problem

A hundred pages of pure text is perhaps 200 KB. Text in a PDF is glyph references plus coordinates, compressed with Flate — it is astonishingly cheap. If your document is large, the weight is in one of these:

• Scanned pages. Each page is a full-resolution photograph. At 600 dpi in colour, a single A4 page can exceed 5 MB. This is the most common cause by a wide margin.
• Embedded photographs at native camera resolution. A 4000 px wide photo placed into a 5 cm box still stores all 4000 pixels.
• Fully embedded fonts. A complete CJK font is megabytes. Subsetting keeps only the glyphs used and usually removes the issue.
• Vector maps and CAD exports. Hundreds of thousands of individual line segments. Genuinely large, and genuinely not compressible without discarding detail.

What each method costs

Downsampling images is the highest-value change. A scan at 600 dpi re-rendered at 150 dpi is roughly a sixteenth of the pixels. For screen reading it looks fine. For print or for a document you may OCR later, 150 dpi is too low — recognition accuracy falls off sharply, so compress after OCR, not before.

Re-encoding images more aggressively as JPEG trades sharpness for size. On photographs this is nearly invisible. On screenshots, diagrams and anything with crisp edges or text it produces the smeared halos everyone recognises. The same reasoning as compressing images for the web applies inside a PDF.

Rasterising whole pages — which is how the compressor here works — re-renders each page at a resolution you choose. It is predictable and it always shrinks a scan-heavy document. But be clear about the cost: a page that contained real text becomes a picture of that text. Selection, search, copy and accessibility are gone. On an already-scanned document you lose nothing, because there was no text to begin with. On a born-digital document you are downgrading it, and the tool says so rather than quietly doing it.

Removing what you do not need costs nothing at all. Delete pages you are not sending in the page organiser, drop attachments, strip metadata. Splitting a 300-page manual into the chapter someone asked for beats compressing all 300 pages — see split.

When not to compress

• Anything going to print. Printers want 300 dpi. Compressing to fit an email and then sending it to a print shop wastes everyone's afternoon.
• Archival or legal copies. Keep the original. Compress a duplicate for circulation.
• Documents you still need to OCR or search. Do the recognition first.
• Files that are already small. Re-encoding a 400 KB text PDF can make it larger and worse. Check the size before assuming there is fat to trim.

The practical order

1. Check what is actually big — is it a scan, or photographs, or neither?
2. Remove pages and attachments you do not need.
3. Run OCR now if you will ever want the text.
4. Downsample to 150 dpi for screen distribution, 300 dpi if print is possible.
5. Open the result and read it before sending. Compression that ruins a figure is not a saving.

All of it happens in your browser, which matters when the oversized file is an invoice run or a medical record — the usual reason a document is both large and confidential.