Getting a PDF Under a 10 MB Upload Limit

The megabytes are in one of two places, and it decides everything

A PDF is a container. Inside it, a page can be described as instructions — draw this glyph here, in this font, stroke this line — or as a picture of a page. The first kind is what a word processor, a spreadsheet or a reporting tool produces. It is astonishingly compact: a page of dense text is a few kilobytes of instructions, and the fonts are shared across the whole document rather than repeated. The second kind is what a scanner or a phone camera produces: a photograph of a sheet of paper, stored as a JPEG, one per page.

The arithmetic that follows is not close. A twelve-page contract exported from a word processor is often under 200 KB in total. The same twelve pages scanned at 300 dpi are twelve photographs of about 8.7 megapixels each, and land somewhere around 12 MB. Nothing about the content changed; the representation did. This is why generic advice to "compress the PDF" is useless without knowing which one you have — the two cases have almost nothing in common except the file extension.

What structural compression actually removes

A PDF is built from numbered objects, and the file carries a cross-reference table saying where each one begins. Editing a document repeatedly, or generating it from a template, leaves that structure fat: objects that nothing points at any more, dictionaries repeated once per page instead of shared, a cross-reference table that has been appended to rather than rewritten. Rewriting the file with object streams packs many small objects into a single compressed stream and replaces the table with a compressed equivalent. That is what the compressor on this site does, and it is genuinely lossless — every glyph, every line, every pixel is bit-for-bit what it was.

The honest limit is the other side of that sentence. Because it is lossless, it cannot touch the image streams — a JPEG inside the PDF comes out as the same JPEG. On a scanned document, the structure is a thin wrapper around a pile of photographs, so removing structural waste removes a rounding error. The tool will tell you it saved almost nothing, and it will be telling the truth: there was almost nothing there to save.

Resolution is quadratic, and that is the whole lever

The page does not change size, so resolution multiplies in two directions at once. Going from 300 dpi to 200 dpi is a factor of two thirds on each axis, and two thirds squared is four ninths: 44.4% of the pixels, 55.6% removed. Going from 300 to 150 is a half on each axis and therefore a quarter of the pixels. This is why resolution is the only lever with real leverage on a scan, and why small adjustments to it produce large changes in file size.

Where to stop is a judgement about who reads the document. Two hundred dpi is the usual floor written into administrative specifications, and at that resolution ordinary body text and signatures remain clean on screen and print acceptably. One hundred and fifty is still readable but starts to soften small print, footnotes and stamp impressions. Below that, characters merge. And if the document will be fed to a text-recognition system rather than read by a person, do not reduce at all: recognition accuracy falls off sharply under 300 dpi, and a rejected form is a worse outcome than a second upload attempt.

The route when you cannot re-scan

Often the scan came from somewhere else and the original is gone. The way through is to take the PDF apart and rebuild it: export each page as an image, reduce those images, and assemble a new PDF from the reduced set. It is three steps rather than one, and every step runs in the browser, so a payslip or a medical form does not travel to do it.

One caution about repeated JPEG encoding, because this route involves it. A JPEG that is decoded and re-encoded loses a little each time, and the loss compounds; it is not visible after one pass at a sensible quality, and it is visible after five. Do the reduction once, at the resolution you actually need, rather than shaving the file down in several attempts. If the first attempt overshoots, go back to the original PDF and start again rather than re-reducing the reduced copy.

When splitting beats shrinking

Read the limit carefully before you degrade anything. Most forms cap each attachment rather than the submission, and a great many accept several. A 14 MB scan that has to stay at 300 dpi because it will be machine-read splits into two 7 MB halves that both upload at full quality. That is a better outcome than a single file at 150 dpi that the recognition system then fails on.

Two smaller cautions worth knowing. Some forms measure the encoded size of an upload rather than the file on disk, which for an email attachment adds roughly a third; if you are sending by mail rather than filling a form, aim for about 7 MB to clear a 10 MB ceiling. And a password-protected PDF cannot be restructured or split without the password, because the tool cannot read the object structure it would have to rewrite — remove the protection first, then work, then re-apply it.

A decision you can make from the file size alone

Put together, the rule fits in a line: kilobytes per page tells you which problem you have, and the problem tells you which tool. Text PDFs respond to structural compression and lose nothing; scans respond only to resolution and lose detail in exchange. Trying the wrong one wastes a few seconds; knowing which is which turns a form rejection into a solved problem, usually on the first attempt.

Frequently asked questions

I compressed it and it barely changed. Is the tool broken?

Almost certainly not — it is telling you the file is a scan. Structural compression is lossless, so it cannot re-encode the images inside the PDF, and on a scanned document the images are essentially the whole file. A result of one or two percent means there was one or two percent of structural waste and nothing else to take. Check the kilobytes per page: over about 300 KB and the answer is resolution, not compression.

Will reducing the resolution make the document legally invalid?

Reducing resolution does not by itself affect validity, but two things can. If the document carries a digital signature, any modification breaks it — the signature covers the exact bytes, so a rebuilt PDF is a different document as far as the signature is concerned, and it will show as invalid. And some procedures specify a minimum resolution in their own rules, commonly 200 or 300 dpi. Check the form's stated requirement before reducing, and if the file is signed, do not rebuild it at all: split it instead, or ask whether the signature can be re-applied.

Why does the same page take more space in colour than in grey?

Because colour stores three channels where grey stores one, and because photographic noise in a colour scan compresses badly. On a black-and-white original — a letter, a form, a contract — scanning in grey rather than colour typically takes off a further third to a half at the same resolution, and loses nothing that was in the document to begin with. Scan in colour only when colour carries meaning: a stamp, a highlighted passage, a chart, an identity document.

Can I just zip the PDF?

It will save almost nothing, because a PDF is already compressed internally — the streams inside it are deflate-compressed, and the images are JPEGs, which is to say already-compressed data. Zipping compressed data typically gains a percent or two. It also usually makes things worse in practice: many forms reject archives outright, and a reviewer who has to unpack an attachment is a reviewer you have annoyed for no gain.

The limit says 10 MB and my file is 9.4 MB, but it is still rejected. Why?

Usually one of three things. The limit may apply to the whole submission rather than to each attachment, so your other files count too. The system may measure the encoded upload rather than the file on disk, which adds roughly a third for a mail attachment. Or the megabyte itself is ambiguous: 10 MB can mean 10,000,000 bytes or 10,485,760, and a file the operating system calls 9.4 MB can be over the smaller of those. Aim for a comfortable margin — around 7 MB against a 10 MB ceiling — rather than trying to land just under it.