How to compress a PDF without losing quality
Truly lossless PDF compression means font subsetting, object streams and removing unused objects, it changes nothing visible and saves 10-30 percent. Anything beyond that downsamples images. For documents you will only read on screen, 150 DPI is visually indistinguishable from 300 and saves about 75 percent.
"Without losing quality" means two different things depending on what is inside your PDF, and most compression advice confuses them. A text document and a scanned document need opposite treatment.
What lossless actually means in a PDF
Lossless is a precise technical claim: decompress the result and you get the original bytes back. In PDF terms, only a handful of operations qualify.
| Operation | Truly lossless | What it costs you |
|---|---|---|
| Font subsetting | Yes | Nothing, unless someone later edits the text and needs a missing glyph |
| Object streams + compressed xref | Yes | Nothing. Requires PDF 1.5 or later |
| Removing unreferenced objects | Yes | Nothing. Removes only data no page points to |
| Re-Flate uncompressed streams | Yes | Nothing |
| Deduplicating repeated images | Yes | Nothing |
| Re-encoding PNG-style images with Flate | Yes | Nothing |
| Downsampling images | No | Pixels, permanently |
| JPEG re-encoding | No | Detail around edges, permanently |
| Flattening forms and annotations | No | Editability of the fields |
| Rasterising pages to images | No | The entire text layer and all searchability |
The last row is the trap. Some "compressors" convert every page to a JPEG and wrap the images back in a PDF. The file gets dramatically smaller and looks fine at 100 percent zoom, but the text is now a picture: unsearchable, unselectable, unusable for screen readers. Always press Ctrl+F on the result and search for a word you can see on screen.
Text documents: lossless is usually enough
A PDF exported from Word, Google Docs or LaTeX is mostly vector text and a few images. There are no pixels to throw away, so the entire saving comes from structure. Realistic expectations:
- A clean 20-page report exported today: 5-15 percent. It was already efficient.
- The same report after twelve rounds of edits and comments: 30-60 percent, because incremental saves left old revisions inside the file.
- A document embedding a full CJK or large icon font: can be dramatic, since subsetting a 12 MB font down to the glyphs actually used is a single huge win.
- A document assembled by merging ten sources: 20-40 percent, because deduplication collapses the ten copies of the same logo into one.
Scans: pick a DPI and accept the trade
A scanned page is a photograph. There is no lossless way to make a photograph meaningfully smaller. The question is how much resolution you can give up before it shows, and the honest answer depends on the destination.
| DPI | Readable on screen | Readable printed | OCR accuracy | Verdict |
|---|---|---|---|---|
| 600 | Yes | Yes, excellent | Excellent | Overkill unless archiving or scanning tiny print |
| 300 | Yes | Yes | Excellent | The safe default when the file will be printed |
| 200 | Yes | Good | Good | Best balance for text scans destined for email |
| 150 | Yes | Acceptable at body-text sizes | Marginal | Fine for reading, risky if you still need to OCR |
| 96 | Small text softens | No | Poor | Previews only |
| 72 | Visibly blurry | No | Fails | Do not use for documents |
Order matters, and so does the level. Run OCR on the high-resolution scan first, because recognition accuracy falls off a cliff below about 200 DPI. Then compress with Light only: Light rewrites the file structure and leaves the text layer alone, while Medium and Strong rasterise every page to a JPEG and rebuild the document, which destroys the text layer you just created. There is a full guide in how to OCR a scanned PDF.
The colour mistake that costs the most
Most people scan black text on white paper in 24-bit colour because that is the scanner default. That stores three channels for a page that has one channel of information. Two better options:
- Greyscale (8-bit), one third of the data, keeps the soft edges that make text look natural, and handles pencil marks, stamps and signatures well. This is the right default for scanned contracts.
- Bitonal (1-bit), one twenty-fourth of the data. With CCITT Group 4 or JBIG2 encoding, a full page of text can be 20-60 KB. Perfect for clean printed text; terrible for anything with photographs, shading or a coloured letterhead.
A safe procedure
- Keep the originalCopy the file before you touch it. Everything below is reversible only if you still have the source.
- Run the lossless pass aloneIn the compressor, compress structurally with no image changes. Note the new size. If it is small enough, stop.
- Decide the destinationScreen only → 150 DPI. Will be printed → 300 DPI. Will be archived or re-OCR-ed → leave the images alone entirely.
- Downsample onceApply image compression to the original, not to an already-compressed copy. Compressing a JPEG twice stacks the artefacts.
- Inspect at 400 percentZoom in on the smallest text and on any diagonal line or fine table rule. Artefacts appear there first, long before they show in body text.
- Check the text layer survivedCtrl+F for a visible word. If it is not found, the tool rasterised your pages, discard the result.
Run a lossless pass first and see how far it gets you.
Compress PDF →If lossless gets you nowhere and downsampling would hurt, the file is simply big. Split it, or send a link. See reducing a PDF for email for the practical limits.
Frequently asked questions
Is there such a thing as truly lossless PDF compression?
Yes. Font subsetting, packing objects into object streams, compressing the cross-reference table and deleting unreferenced objects are all bit-exact lossless. They typically save 10-30 percent. Anything advertising 90 percent savings on a text PDF is not lossless.
What DPI should I compress a PDF to?
150 DPI for on-screen reading, 200 DPI for scanned text you may still print, 300 DPI for anything going to a printer properly. Below 150 the small text starts to soften; below 100 it becomes visibly blurry.
Will compressing a PDF break the OCR text?
It depends on the level. Light is structural and lossless, so the text layer is untouched. Medium and Strong rasterise each page to an image and rebuild the file, which removes every text object, the result looks identical but nothing is searchable any more. OCR first, then compress with Light, and check with Ctrl+F before you delete the original.
Why does my compressed PDF look blurry when I zoom in?
The images were downsampled below what your zoom level needs. A 150 DPI page looks perfect at 100 percent and soft at 300 percent. If you need to zoom, keep 300 DPI, or go back to the original file.
Does compressing a PDF twice make it smaller?
A second lossless pass gains nothing. A second lossy pass makes it smaller but stacks JPEG artefacts, and the damage compounds visibly. Always compress the original once at the setting you actually need.