How to Reduce PDF File Size Without Losing Quality

The honest version of this topic starts with a correction. "Without losing quality" is achievable for some PDFs and impossible for others, and which category your file falls into is decided before you open any compression tool. So the first job is not compressing. It is finding out what is actually taking up the space.

First, work out why your file is big

Divide the file size by the page count. That single number tells you most of what you need to know.

• Under about 50 KB per page. This is a text document with subset fonts. It is already efficient. Compression will save you almost nothing, and any tool promising 80% off is going to achieve it by wrecking something.
• Roughly 100 to 500 KB per page. Mixed content: text plus a few images, charts, or a logo on every page. There is room to improve, mostly in the images.
• Over 1 MB per page. Almost certainly a scan, or slides exported with full-resolution photographs. This is where real reductions live, and it is the only case where aggressive compression is the right answer.

A second check takes five seconds: try to select text on a page. If the cursor highlights words, you have a text PDF. If it selects nothing, or draws a box across the whole page, every page is an image and you are dealing with a scan. These two file types need opposite treatment, and confusing them is the most common mistake people make.

What is actually lossless, and what it saves you

There are genuinely lossless ways to shrink a PDF. They just do not save much, which is why nobody advertises them.

• Font subsetting. Embedding only the glyphs the document uses instead of the whole typeface. A full Unicode font can be several megabytes; the subset for a report using 120 distinct characters is a fraction of that. Most modern export pipelines already do this, so the saving is usually already banked.
• Object stream compression and duplicate removal. Packing the file's internal objects into compressed streams and collapsing repeated resources, such as the same logo embedded fifty times. Genuinely lossless, and typically worth single-digit percentages on a file that was not badly generated in the first place.
• Dropping unused parts. Embedded page thumbnails, old incremental-save revisions, unused form field definitions, orphaned annotations. Occasionally this finds a surprising amount, particularly in a PDF that has been edited and re-saved many times over years.

Everything else that shrinks a PDF meaningfully is lossy. Downsampling images throws away pixels. Re-encoding them at a lower JPEG quality throws away detail. Converting to grayscale throws away colour. These are the levers that actually move a 40 MB scan, and every one of them costs you something. The only real question is whether the cost is visible at the size the document will be read.

Exactly what RaptorPDF's compressor does

We would rather describe this precisely than call it intelligent. Our compressor renders each page with pdf.js and re-embeds the result as a JPEG image. That is the whole mechanism. The three settings are:

When this approach is wrong

Rasterising a page is a blunt instrument, and on the wrong document it does real damage. Before you run our compressor on a text PDF, know what happens:

• Your text stops being text. It cannot be selected, copied, or found with Ctrl+F. If you send a contract to someone who needs to quote a clause, you have just made their job much harder.
• Links die. Clickable URLs, cross-references, and the table of contents become pictures of themselves.
• Screen readers go silent. An image of a page is inaccessible. If the document has any obligation to be readable by assistive technology, rasterising it is not an option.
• The file can get bigger. A page of vector text might occupy 20 KB. A 101 dpi JPEG photograph of that same page will not. On text-only documents, compression by rasterisation regularly produces a larger file than it started with.

For a text document that is genuinely too large, the fix is upstream. Go back to the original Word document, Google Doc, or Keynote file, downsample the images inside it, and export again. Word and Pages both have a reduced-size PDF export option that downsamples images while leaving text as text, which is the thing a page-rasterising compressor cannot do. If the source file is gone, a desktop tool with a real image-level optimiser is the correct purchase; we do not have one.

Compressing for email, with the arithmetic

Gmail caps messages at 25 MB, and this is where people get caught out. Attachments are base64 encoded for transport, which inflates them by about a third. A 19 MB PDF becomes roughly 25.3 MB on the wire and bounces. Aim for 18 MB or less if you want a 25 MB limit to hold, and lower if you do not know the recipient's server configuration, because corporate mail gateways are often stricter than the consumer services.

If the file will not get under the limit without becoming unreadable, stop compressing and change the delivery method. Split the document into parts, or share a link. A split by page range is lossless, because it copies the original page objects rather than re-rendering them, so two halves of a scan are exactly as sharp as the original was.

Check the result before you send it

1. Try to select text. If you cannot, the page was rasterised. Decide whether that is acceptable for this document.
2. Zoom to 100% and read the smallest text on the page. Not the heading. The footnote, the figure caption, the signature block.
3. Check one photograph closely. JPEG at quality 0.50 produces blocking artefacts around high-contrast edges, and faces show it first.
4. If it will be printed, print one page. Screens forgive 100 dpi. Paper does not.

The rule worth remembering

Scans compress well because they are photographs of paper and contain far more data than the information on the page requires. Text PDFs compress badly because they are already close to the minimum size their content allows. If your file is a scan, compress aggressively and use grayscale. If it is text and still too big, the problem is an oversized image inside it, and you fix that where the document was written, not here.