How to reduce PDF file size

The three things that make a PDF big

A PDF is heavy for one of three reasons. Find yours before you set a number, because the reason decides how much weight comes off.

1. Scanned pages. Every page is a photograph of paper, so the file weighs what the photographs weigh, and the text on them is not text at all.
2. Placed images. A report with eight camera photos in it carries eight camera photos, at a resolution no printer and no screen will ever use.
3. Embedded fonts and leftover structure. A font program usually ships every glyph it was designed with, and a document usually draws a few hundred of them. Old revisions and orphaned objects sit in the file after edits and never get cleaned out. This one is real but it is rarely the whole story.

Tell them apart in five seconds. Open the file and try to select a line of text with the cursor. If the text highlights, the pages are not scans and the weight is in the pictures or the fonts. If nothing highlights, the page is an image and the scan is the file.

What the tool does about each one

The compress tool runs four methods against your file at the same time and keeps whichever result is smallest and still passes its check. The structural method rebuilds the document into a fresh file, dropping orphaned objects and old revisions. The image method re-encodes the placed images at 150 dpi with mozjpeg at quality 75. The font method subsets each embedded TrueType down to the glyphs the pages actually draw.

The fourth is the raster method, and it is the only one that touches the page itself. It renders each page at 150 dpi and redraws the page's own text invisibly on top, so the words stay selectable and searchable even though the picture underneath is now an image. It starts only when no other method reached your number, and it gives up early when its own projection cannot beat a candidate another method already holds.

On the published run the raster method won 52 of 120 cases, the structural method 38, the font method 17 and the image method 13. So there is no single answer to what shrinks a PDF. It depends on which of the three problems your file has.

Type a size, not a quality level

Most tools give you a slider named low, medium and high, which tells you nothing about the file you get. This one takes a number in KB and works toward it. One KB is 1,024 bytes here, so a 200 KB target means the output lands under 204,800 bytes.

If an upload form counts a kilobyte as a flat thousand bytes, its cap is 4,800 bytes tighter than that. When a file that reads as the right size still gets refused, set the target about 3% lower and run it again.

How much usually comes off

Across the 120 public cases the median reduction was 59.2% of the source bytes and the top tenth reached 88.1%. The least compressible file gave back 3.8%. No output came back larger than the file it started from.

Document kind decides most of it. Scans met their size target on 16 of 18 cases, 88.9%, because image quality is a dial with a long travel. Mixed documents met theirs on 29 of 52. Text documents met theirs on 10 of 50, 20.0%, because their bytes are fonts and structure and there is far less spare weight to give back. A guide that promises a fixed percentage on any file is guessing.

Check the result before you send it

Nothing ships unread. Every candidate is opened again and its text is compared against the source, and any candidate that lost more than 2% of the non-space characters is thrown away rather than offered to you. On the public run, 102 documents carried text, all 102 came back at 100% character retention, and no candidate was refused.

That rule is why the file you download is sometimes bigger than the number you asked for. When no method reached the target, you get the smallest file that passed the check plus the size it actually reached, and you decide what to do about the gap.

From here, the quality guide covers which methods cost you nothing and which ones cost you detail. The compress hub lists the rest of the cluster, including phones, very large files and merged documents.