A PDF made from a word processor is usually small: text and fonts compress well. A large PDF almost always contains images, most often pages scanned or photographed at photo quality, sometimes the same picture stored many times. Fix the images and the file shrinks.
01Where the bytes go
Inside a PDF, each page is a short program that draws text and shapes, plus the resources it uses: fonts, images and, for forms, field data. Text itself is tiny. Fonts are embedded once per file. Images are where size explodes, because a scanned page is a picture of every dot on the paper.
| File | Size | Pages | Images | Embedded fonts | Other streams |
|---|---|---|---|---|---|
| IRS W-9[2] | 140,815 B | 6 | 0 B | 17,737 B (12%) | 98,124 B (69%) |
| IRS W-4 2026[3] | 208,845 B | 5 | 1,743 B (<1%) | 20,353 B (9%) | 159,659 B (76%) |
| USCIS I-765[4] | 467,424 B | 7 | 118,694 B (25%) | 15,310 B (3%) | 272,102 B (58%) |
| USCIS I-765 instructions[5] | 580,406 B | 25 | 0 B | 211,104 B (36%) | 262,267 B (45%) |
Method: pypdf 6.6.2, summing the stored (compressed) length of every stream by type: image XObjects, embedded font files, and everything else (page content, form and XFA data, metadata). Percentages are of the file size; the remainder is object structure and the cross-reference table.
None of these born-digital forms is big: the largest is under 600 KB for 25 pages. The I-765 is about a quarter images (nine image objects across seven pages); the instructions are about a third fonts, because they embed six font files. Neither would ever trouble a 12 MB upload limit.
02Scans: the usual culprit
Scanning settings change size by an order of magnitude. The Northern District of California tested one document: “Our testing shows that a 317 page document scanned: -- bitonal, at 300 ppi, was 23.86 MB and sufficiently legible, whereas -- gray scale, at 400 ppi, was 374.47 MB.”[1] Same pages, about sixteen times the bytes. The court’s advice for a file that is too big: “Rescan your document at LOW resolution or "draft" quality and you'll create a much smaller file size.”[1]
It also asks the question worth asking first: “If your .pdf file was scanned, did it really have to be scanned?” A document that exists as a file should be converted to PDF directly, not printed and scanned.[1]
For text documents, resolution and color depth are the two dials that matter. A page of black type does not need 16 million colors.
Our reading of [1]
03Other causes, and the fix for each
- Photos pasted at camera resolution. A phone photo inserted into a document keeps all its pixels. Fix: downsample images to the size they are shown at, then re-encode as JPEG.
- The same image stored many times. A logo or letterhead on every page should be stored once and reused; some tools embed a copy per page. Fix: de-duplicate identical images.
- Many embedded fonts. Each typeface (and often each weight) is embedded separately; the I-765 instructions above carry six font files. Fix: subset fonts to the characters used.
- Merged files carrying their history. Combining many PDFs can bring along unused objects and metadata. Fix: rewrite the file with a structural optimizer.
04Measure your own file
Why is my PDF so large? reads the file in your browser and prints its byte budget (images, fonts, duplicates and other) with a fix next to each finding. When you know your target, the compressor works toward an exact size, and the upload limits list says what “small enough” means for USCIS, courts, email and AI assistants.
Figures in the table are our own measurements of the named files on the date shown; other files will differ.