This is a condensed version of a study first published on the EasyPDF blog, with the full country surveys and every source link.

Ask people why they compress a PDF and most will describe a huge file: a 40 MB scan, a brochure full of photos, a report that will not go through email. Our production data says something different. The PDFs people actually bring to an online tool are small and short. Half of them weigh less than 273 KB. Half of them are a single page. And yet compression accounts for 49.4% of everything our users do, and most of that compression happens on a phone.

The data: file size and page count for 2,023 documents processed between 5 February and 6 September 2026, and tool usage for 23,379 sessions over 12 months. The short version: the "1 MB problem" is not a big-file problem. It is an upload-limit problem.

What a typical PDF looks like

The median PDF in our sample weighs 273 KB and has 2 pages. The mean is 2.20 MB, eight times the median: the distribution has a long tail rather than a fat middle.

Percentile File size Pages
p10 42 KB 1
p25 103 KB 1
p50 (median) 273 KB 2
p75 698 KB 4
p90 2.32 MB 11
p95 5.27 MB 27
p99 21.1 MB 127

81.8% of PDFs are under 1 MB and 68.0% are under 500 KB. 49.0% have exactly one page. Dividing size by pages gives a median of 128 KB per page. What makes a PDF heavy is scanned pages, not length: one page scanned at 300 DPI typically weighs 2 to 5 MB on its own.

Compression is half of everything people do

Across 23,379 tool uses, compress represents 49.4% of sessions. Editing text comes second at 30.7%, then conversion at 7.8%, translation at 5.5% and OCR at 2.5%. Merge (1.8%) and split (1.5%) are marginal.

Four out of five files are already under 1 MB, yet one out of two sessions is a compression. If people compressed because files were large, compression would be a niche tool used by the top decile. It is the opposite.

The explanation: upload limits, not big files

The missing variable is the form on the other side. Online portals cap the size of the documents they accept, and those caps are far below what a PDF weighs as soon as it contains an image.

  • Indonesia: civil-service recruitment portals (CPNS, SSCASN) cap most documents at 1 MB or less.
  • India: exam forms such as NEET, JEE and UPSC cap scanned PDFs at 300 KB, photos and signatures at a few tens of KB.
  • Europe: many public and university portals sit at 2 to 5 MB per attachment.

The p75 file at 698 KB fits a 1 MB limit, but not a 500 KB one. The p90 file at 2.32 MB fails 1 MB and 2 MB limits. For most users, compression is not "make this file smaller". It is "make this file fit under the number printed on the form". That is why the compress pages are organised by target size: 1 MB, 2 MB, 5 MB.

Five countries, two scales

We recorded the cap published by the official help pages of 162 portals in France, the US, the UK, Germany, Spain and India. Germany, Spain, the US and the UK sit at a median of 10 MB, France at 5 MB. India works in kilobytes: 500 KB across its firm caps and 300 KB on exam forms. In France, the spread goes from 1 MB (vehicle registration at the ANTS) to 200 MB, and 7 portals out of 28 publish no limit at all: users find out when the upload fails.

Compression happens on phones

Among 6,566 compress sessions with a known device type over 90 days, 3,770 came from a phone, 2,722 from a desktop and 74 from a tablet: 58% mobile.

What this means if you build or publish PDFs

  • Budget 128 KB per page. A 4-page document at that budget lands around 500 KB and passes almost every cap except the Indian exam ones.
  • Scanning is the size driver, not length. Scan at 150 to 200 DPI in greyscale, or run OCR so a text layer replaces the pixels.
  • Know the target before you compress. Compress to just under the cap, not to the smallest possible size, to keep the maximum quality the limit allows.
  • If you build an upload form, state the cap in KB next to the field, and remember the person filling it is probably on a phone.

Open data

The portal caps are published as an open dataset under CC BY 4.0, with the source URL of every figure: pdf-upload-caps on GitHub. A searchable version is on the PDF size limits page.

Methodology and bias. The sample is EasyPDF users, so it over-represents documents that must fit a constraint and under-represents PDFs that are simply read or archived. Nothing here is derived from document contents: only file size, page count and anonymous usage events. Full methodology in the original study.