Your PDF Is Too Big to Upload — Here’s the Fix

Upload forms rarely tell you before they reject you, and the numbers are not round. These are published limits, quoted from the source:

Sources, in order: USCIS, Tips for Filing Forms Online; GOV.UK, EU Settlement Scheme evidence; N.D. Cal. e-filing limitations and C.D. Cal. notice; Common App recommender support; LinkedIn Help.

Two patterns fall out of that list. Court systems set their limits per district, so PACER’s own answer is “each court sets a limit” and points you at a lookup tool. And several portals cap the number of documents as well as the bytes, which changes the strategy: if you are allowed ten files at 6 MB each, splitting is a legitimate move. If you are allowed one, it is not.

Email is a different problem with a different mechanism, because attachments are base64-encoded in transit and grow roughly a third on the way. That is covered in getting a PDF under the email limit.

One command answers this, and it decides which route works. Install poppler with brew install poppler, then:

pdfimages -list yourfile.pdf

You get a row per embedded image with its pixel dimensions and its effective resolution on the page. Three outcomes:

• Rows with high PPI numbers. Photographs placed far larger than the page needs. This is the common case and it compresses beautifully. One of our test files carried images at 1,719 and 2,824 PPI.
• Rows at 200–300 PPI, one per page. A scan. Compressible, but resolution is the only lever and the text is made of pixels. See compressing a scanned PDF without killing OCR.
• No rows at all. A text-only PDF. Nothing here will help much, and some tools will make it worse. Skip to the mistakes section.

Measured on three files that fail for different reasons: a 34.8 MB twelve-page design export with photographs placed at camera resolution, a 16.8 MB eight-page brochure with full-bleed images, and a 5.02 MB three-page 300 PPI scan.

The useful reading is not the smallest number in each column. It is that almost everything clears a 2 MB cap, which is the tightest limit in the list above. The real question is how much quality you are handing over to get there, and that is set by the resolution the preset leaves behind rather than by the byte count.

A practical rule that has served us well: aim for 70% of the stated limit. Portals count the encoded upload, some of them re-wrap the file, and there is nothing worse than a 11.8 MB file against a 12 MB cap failing for a reason the form will not explain. If the cap is 2 MB, target 1.4 MB.

For how each preset changes what is on the page rather than what is in the byte count, the preset breakdown goes through them one at a time.

All three were measured on this run, and all three are things people try first.

Running Preview’s Reduce File Size on a text-only PDF. Our three-page, 87,788-byte text document came out of it at 109,049 bytes, 24.2% larger. The filter only shrinks images, and it rewrites the whole document through Quartz’s PDF writer on the way, which is more verbose than whatever produced the file. Ghostscript took the same file to 67,396 bytes, down 23.2%. Full autopsy in why Preview’s Reduce File Size ruins PDFs.

Splitting a PDF whose pages share images. This one surprises people. Splitting our twelve-page layout into three four-page files produced 91.5 MB of output from a 34.8 MB input, an increase of 162.8%, because each part carries its own copy of every shared image object. On the scan, where each page has a unique image, splitting into three came to 5,016,808 bytes against a 5,017,560-byte source, a difference of nothing. Check pdfimages -list for repeated object IDs before you reach for the knife.

Using a “print quality” preset on a 300 PPI scan. Ghostscript’s /printer and /prepress target 300 dpi and only act on images above 450 PPI. Feed them a scan that is already at exactly 300 and they do nothing: our 5,017,560-byte file came back at 5,019,305 bytes, and the command reported success. A silent no-op is the hardest failure to debug, and it is why the same preset that saves 95% on one file saves nothing on another.

Sometimes the honest answer is that the file cannot be that small and still be the document it needs to be. In rough order of how much they cost you:

• Convert to grayscale before you drop resolution. Our 300 PPI color scan went from 5,017,560 to 1,720,854 bytes, down 65.7%, with every pixel of resolution intact. For black ink on white paper this is nearly free, and it is the move most people skip.
• Split it, if the portal allows more than one file. Several do, and several cap the count rather than the size. Check for shared images first: on a scan it is lossless, on a shared-image layout it can multiply the total.
• Rebuild from the source instead of squeezing the export. If the PDF came out of a layout application, the fix belongs upstream, in the export dialog rather than in a compressor. Why your InDesign PDF export is enormous covers exactly that.
• Re-scan at a lower resolution. Slow, but a scan captured at 200 PPI grayscale is cleaner than a 600 PPI color scan crushed into the same byte count, because downsampling amplifies scanner noise.
• Send a link, but only if the portal accepts one. Most government and court systems do not, and an emailed link to a court filing is not a filing. Check before you assume.

What we would not do is pass a confidential document through a free web compressor to save four minutes. Visa paperwork, medical records, court exhibits and signed contracts are exactly the documents where the upload is the problem, not the file size.

Your PDF is text-only. All five of our presets produced the identical 79,132 bytes from the 87,788-byte text document, a 9.9% saving from structural repacking alone, because there were no images to work on. Ghostscript managed 23.2% on the same file and it is free. If your PDFs are word-processor exports, install it and write a one-line function: the terminal guide has the command.

It is one file and you are in a hurry. Preview’s Reduce File Size took the 16.8 MB brochure to 4.05 MB in about a second and a half, free, with nothing installed. For one upload against a 12 MB cap that is the end of the problem.

You need the PDF edited, redacted, or OCR’d. We compress and convert. We do not edit PDFs, and a redaction you draw a black box over in the wrong tool is not a redaction.

Where a local app earns its place: several files at once, documents that must not leave your machine, and a ladder of quality settings instead of one button. Smol is $29 once, with no per-file ceiling and no account. If you want to see how it sits against the rest of the field, including where it loses, we ranked the Mac PDF compressors.