How to compress a PDF without losing quality

A PDF is not one kind of file. A text-only contract exported from Word, a 40-page scan of that same contract, and a photo-heavy brochure are structurally different documents, and they compress completely differently. The single most useful thing to know before compressing a PDF is where the bytes are — because in almost every oversized PDF, the answer is: in the images.

Why PDFs get large

A PDF page is a program of drawing instructions plus a set of embedded resources. The instructions themselves — text positioning, vector paths — are tiny and already Flate-compressed (the same DEFLATE used in ZIP and PNG). The resources are where size accumulates:

• Embedded images dominate. A full A4 page scanned at 300 DPI is roughly 2480×3500 pixels — about 8.7 million pixels per page, before compression. One page of scan often outweighs a hundred pages of native text.
• Fonts. Each embedded font adds tens to hundreds of kilobytes; full (non-subsetted) fonts, especially CJK ones, can add several megabytes. Subsetting — embedding only the glyphs actually used — fixes this losslessly.
• Duplicate resources. Merging PDFs or generating them page-by-page can embed the same logo, background, or font once per page. Deduplication collapses these to a single shared object.
• Incremental-save baggage. PDF supports appending edits to the end of the file rather than rewriting it. A heavily edited document can carry every previous version of itself. A full re-save ("save as") discards the history.

Adobe Acrobat Pro’s “Audit space usage” (inside the PDF Optimizer) shows this breakdown per document. If images are under 20% of the file, aggressive image recompression won’t help much — look at fonts and structure instead.

Lossless vs lossy PDF compression

“Without losing quality” has a strict reading and a practical one.

Strictly lossless operations change no rendered pixel: re-compressing streams with Flate, converting to object streams (a PDF 1.5+ structure that packs many small objects into one compressed stream), deduplicating resources, subsetting fonts, stripping metadata and embedded thumbnails, and dropping incremental-save history. On a digital-born document these typically recover 5–30% — worthwhile, but rarely transformative.

Practically losslessmeans lossy image recompression tuned so the difference isn’t visible at the sizes people actually view the document. This is where the big reductions live, and it hinges on two operations applied to each embedded image:

• Downsampling— reducing the image’s pixel dimensions to match its effective resolution on the page. An image placed at 3 inches wide needs no more than ~450 pixels of width for a 150 DPI target. Halving resolution quarters the pixel count; going from 600 to 150 DPI cuts it 16×.
• Recompression — re-encoding the pixels with a stronger codec or lower quality: JPEG (DCTDecode) for photos and scans, JBIG2 for black-and-white scanned text, JPEG 2000 (JPXDecode) in some workflows. The same quality-vs-artifacts trade-offs from image format selection apply, just inside the PDF container.

Quality visibly suffers when downsampling drops below what the reading context needs (scanned text gets fuzzy below ~150 DPI; fine print and small handwriting degrade first) or when JPEG quality is pushed low enough that ringing artifacts appear around high-contrast edges — which is exactly what scanned text is made of. Photographs tolerate low JPEG quality far better than text-on-white scans do.

Method 1: in the browser

Our PDF compressorruns entirely in your browser — the file is processed locally and never uploaded to a server, which matters for contracts, medical records, and anything else you wouldn’t email to a stranger. It applies the standard pipeline: downsample embedded images to a target resolution, recompress them as JPEG at a chosen quality, and rewrite the document with compressed streams. Pair it with PDF splitif a single attachment limit still can’t be met, or PDF merge to combine parts after the fact.

Method 2: macOS Preview

Open the PDF in Preview, then File → Export, set format to PDF, and choose the Reduce File Size Quartz filter. This downsamples and JPEG-recompresses images in one step. The stock filter is aggressive — on scans it can make text noticeably soft. If the output looks too degraded, you can create a custom Quartz filter in the ColorSync Utility with a gentler resolution floor and higher JPEG quality. Always Export, not Save — export writes a new file and leaves the original untouched.

Method 3: Ghostscript

Ghostscript’s pdfwrite device is the standard command-line tool for this, via the -dPDFSETTINGS distiller presets:

• /screen — 72 DPI images, low JPEG quality. Smallest output; scanned text suffers visibly.
• /ebook — 150 DPI, medium quality. The sweet spot for email and on-screen reading.
• /printer — 300 DPI, higher quality. For documents that will be printed.
• /prepress — 300 DPI, color-preserving, minimal loss. Largest output of the four.

A typical invocation:

gs -sDEVICE=pdfwrite -dCompatibilityLevel=1.6 -dPDFSETTINGS=/ebook -dNOPAUSE -dBATCH -sOutputFile=out.pdf in.pdf

Each preset is shorthand for a bundle of parameters (ColorImageResolution, ColorImageDownsampleType, JPEG quality, font subsetting) that you can override individually — see the Ghostscript VectorDevices documentation in the sources. Two caveats: Ghostscript re-interprets and rewrites the whole document, so interactive features (form fields, signatures, tagged-PDF accessibility structure) may be flattened or dropped, and a digital signature is invalidated by any rewrite, by design.

Method 4: Adobe Acrobat Pro

Acrobat offers two levels. File → Reduce File Size is the one-click version. The PDF Optimizer (Save as Other → Optimized PDF) exposes the full controls: per-image-type downsampling targets and codecs, font unembedding/subsetting, discarding embedded thumbnails and metadata, and the space-usage audit mentioned above. It is the most precise of the four methods and the only one here that can selectively unembed fonts or apply JBIG2 to monochrome scans with a UI. The trade-off is the subscription.

Realistic expectations by document type

No honest tool promises a fixed percentage, because the outcome depends on what the PDF contains and how inefficient it currently is:

• Scanned documents (image-per-page): the biggest wins. A 600 DPI color scan downsampled to 150 DPI and JPEG-recompressed routinely shrinks 5–20×, because you are cutting pixel count 16× before compression even starts. Monochrome text scans recompressed with JBIG2 can go further still.
• Digital-born text documents: already efficient. Expect 10–50%, mostly from font subsetting, object streams, and structural cleanup. If such a file is inexplicably huge, suspect duplicate resources or incremental-save history rather than images.
• Photo-heavy reports and brochures:somewhere in between — typically 2–5× when print-resolution photos are downsampled for screen use. The images were usually placed at far higher resolution than their printed or displayed size requires.
• Already-compressed PDFs:near zero. Running a compressor twice yields little on the second pass — and re-JPEG-ing already-JPEG-ed images accumulates artifacts without saving much. If the first pass wasn’t small enough, redo it with more aggressive settings from the original, don’t chain passes.

Which method, when

• Confidential file, no installs: a browser-local tool like ours — verify the tool actually processes locally.
• On a Mac, one-off:Preview’s export filter; check the output for soft text.
• Batch jobs, servers, reproducibility: Ghostscript with an explicit preset — scriptable and deterministic, but flattens interactive features.
• Surgical control or JBIG2 on scans: Acrobat Pro’s PDF Optimizer.

And in every case: keep the original. Image recompression is irreversible, and the version you shrank for one email’s attachment limit shouldn’t become the only copy you have.