Why Is My PDF So Large?

Start with file size per page

Write down the total file size and page count, then divide one by the other. This does not prove what is inside the PDF, but it quickly separates a short document carrying megabytes on every page from a long document whose total grew through repetition. Our two-page July scan averaged about 2.61 MB per page; the 240-page dense-text file averaged about 16.4 KB per page. Both started in the multi-megabyte range, yet they needed different explanations. The six measured examples below use binary units for the July files and retain the recorded June figures. Per-page values are simple calculations from those measured sizes and page counts.

Six diagnostic examples. Per-page figures are approximate and are shown as clues, not thresholds.

Photographs are often the obvious weight

A PDF can contain photographs at a much higher resolution than the page ever displays. A four-page report may look restrained on screen while storing substantial image data behind each frame. Our June photo-heavy report measured 11.06 MB, or about 2.77 MB per page. Image down-sampling gave the compressor room to work: the same report reached 455 KB at the 500KB target and 96 KB at 100KB. The saving is real, but so is the trade-off. At a strict target the tool has to remove image detail. If the photographs are evidence, diagrams, products or faces that someone must inspect closely, re-exporting the source document with sensibly sized images is safer than repeatedly compressing the finished PDF.

A scan makes every page an image

A scanned page may show nothing but typed words, yet the PDF stores a grid of pixels rather than those words as characters. That is why a three-page colour letter in the June test occupied 7.16 MB, about 2.39 MB per page, and why the new two-page scanned-style sample occupied 5.22 MB (5,469,469 B), about 2.61 MB per page. Both had abundant raster data to discard. The July sample reached 93 KB (94,894 B) at 100KB, then settled at 299 KB (306,377 B) for the 500KB, 1MB and 2MB targets. For a scan, the best source fixes are usually fewer pixels, less colour where colour carries no information, tighter cropping and fewer unnecessary pages.

Embedded fonts can add weight without looking unusual

PDFs embed font data so the document looks correct on a device that does not own those typefaces. A normal export usually includes only the characters it needs, but some workflows embed larger font programmes, several weights, or repeated subsets. The page still looks like ordinary text, so this cause is easy to miss. Try selecting a sentence: if the text highlights cleanly and the file is heavier than its images and page count seem to justify, the export settings and fonts deserve attention. The safe fix is to return to the source application and export again with proper font subsetting. Deleting font objects from a finished PDF by hand risks changed characters and broken layout; it is not a sensible emergency shortcut.

Text streams can be dense even when there are no images

“Text PDFs are tiny” describes a lean word-processor export, not every file containing selectable text. Our June control made from a word processor was 5 KB across three pages, about 1.7 KB per page, and passed through unchanged. The July dense-text sample was deliberately different: 240 pages of repeated, uncompressed content-stream instructions produced a 3.84 MB (4,023,756 B) input, around 16.4 KB per page. Compression reorganised those streams to 135 KB, but it could not remove the text or pages, so the 100KB run stopped at a 135 KB best effort (138,689 B). Page instructions and vector drawings have a structural floor; unlike photographs, they do not offer a quality slider that can keep discarding pixels.

Already-optimised files may have nothing useful left to remove

The 736 B July control was already below every available target. Each run returned it unchanged. The June 5 KB text export did the same. A small or previously optimised PDF can therefore appear unresponsive for the correct reason: there is no redundant image data or loose stream encoding worth touching. Running it through several services will not create a meaningful saving, and repeated conversion can introduce damage that the original did not have. Check the receiving limit again. If the file is already under it, stop. If it is just above a much stricter cap, the next move may be structural, such as removing an optional page, rather than another generic compression pass.

Use three checks before choosing a fix

First, calculate the rough weight per page. Multiple megabytes per page usually points towards scans or large photographs; a modest per-page figure spread across hundreds of pages points towards accumulated text, vector or font data. Second, try to select individual words. If the whole page behaves like one picture, treat it as a scan. Third, zoom into photographs, stamps and line drawings. A mixed PDF may contain selectable text on one page and a heavy scan on the next, so no single label describes the whole file. These checks do not require specialist software. They simply stop you applying an image remedy to a structural problem, or blaming fonts when two full-resolution photographs are plainly carrying the weight.

Mixed documents need a page-by-page diagnosis

Many application bundles are hybrids: a small exported form, several phone photographs, a scanned signature page and a letter from another system, all joined into one PDF. The total size hides which page is responsible. Scroll through thumbnails and compare visually dense pages with plain ones before changing the whole file. If one photograph or scan is carrying most of the weight, replace or rescan that page rather than down-sampling crisp text everywhere else. If every page is light but the document is very long, page count and structural data are the likelier floor. A targeted repair keeps more evidence intact than applying the harshest setting to the complete bundle.

How we measured

We built three synthetic PDFs to isolate scan-like, dense-text and already-optimised behaviour, then sent each target to the live PDF Size Reducer endpoint and recorded the returned byte sizes.

• Measured and documented by PDF Size Reducer on 20 July 2026 using the live tool.

Which fix applies to which cause

Use the least destructive fix that addresses the material actually occupying the file. Work from the source document when you have it, keep an untouched original, and reopen the output before submitting it. The table is a decision aid rather than a guarantee: real PDFs often combine several causes, so start with the strongest clue and reassess after one careful pass.

Match the observed cause to a source-level fix before applying stronger compression.

Common questions