Make a PDF smaller, with the method that fits your file

What takes the space in a PDF

Open a big PDF and you find 4 kinds of content: text, fonts, images, and the structure that holds the pages together. They weigh nothing alike.

The text itself is nearly free. A page of prose is stored as a short compressed stream, a few KB at most. Fonts cost a little more, and the structure (the page tree, the cross-reference table, the metadata) is small housekeeping. Images are everything else, and in a big PDF they are nearly all of the bytes: in our 10-page scan of a lease, the page pictures are almost the whole 17.6 MB.

This is why page count tells you almost nothing about size. A long contract typed in a word processor can weigh less than 1 scanned page, because the scan stores the page as one big photograph while the typed version stores only the characters and a font. It is also why the same tool produces wildly different results on different files: the ratio belongs to the file, not the tool.

So before you squeeze, know what you are squeezing. If the file came from a scanner, a phone camera or a design tool, it is images and it will shrink hard. If it came out of a word processor with no pictures in it, it is already close to its floor. When you are not sure, open it in the inspector and look at what is inside.

Pick the lever that fits your file

There are 4 levers that make a PDF smaller, and each fits a different file.

Your file	The lever that works	Where
Scans, photos, design exports	Re-encode the images at a resolution that fits the destination	The compressor below
Colour photos that do not need colour	Go grayscale, then compress, for a modest extra saving	Grayscale PDF
Only some pages matter to the receiver	Send those pages, not the file	Extract pages
Still over the cap at readable quality	Split into parts under the limit	Split PDF

The first lever is the big one, and it is the one this page’s tool pulls. The others matter when re-encoding alone is not enough, or when the honest answer is that your file should not be compressed at all; more on that below.

Grayscale helps less than its reputation. On our 10-page colour scan of a black-on-white lease, grayscale on its own saved 12 %, and after the Email preset it saved nothing: once a scan is compressed, its colour costs little. Where it does pay is colour photos you do not need in colour (5 phone photos came to 482 KB grey against 558 KB in colour on the Email preset), and on pages that must print predictably. Colour that means something (a stamp, a signature in blue ink, a chart keyed by colour, a highlighted clause) does not survive it, so look at the pages first. The grayscale guide has the numbers and the cases where colour must stay.

Fewer pages is the lever people forget while fighting with quality sliders. If the receiver needs 5 pages of a 60-page report, the smallest possible file is not a heavily compressed report, it is those 5 pages at full quality. Extracting beats compressing whenever the document is bigger than the job it has to do.

Compress it here

Drop the file, pick a preset, read the result. The presets set the resolution your images are resampled to, which is the single strongest size lever there is:

Preset	Images become	Pick it for
Email	110 dpi	Attachments, portal uploads, anything read on a screen
Web	150 dpi	Sharing and screens where the reader may zoom in
Print	300 dpi	Paper, archives, anything that must survive a close look

Under the hood, the compressor does more than resample:

1. Photos and scans are re-encoded as JPEG at the preset’s resolution. This is where almost all the saving comes from, and it is lossy: pixels change, by design, to a degree that fits how the file will be read.
2. Flat graphics stay lossless. Logos, diagrams and line art are not photos; JPEG would smear their edges, so they are kept in a lossless encoding and stay crisp.
3. Duplicate images are merged. A letterhead or logo repeated on all pages is stored once instead of once per page.
4. Hidden weight is dropped. Old page thumbnails are removed and objects nothing references any more are swept out. These are bytes you were carrying for no reader at all.
5. Text is never touched. It is not an image, so it is never re-encoded. It stays sharp at any zoom, and it stays selectable and searchable in the result.

That last point answers the most common worry up front: compressing a PDF here does not make the writing blurry. If your text looks soft after compression, it was a scan to begin with, which means the “text” was a photograph of text all along. There is a guide on compressing scanned PDFs without wrecking the text, and if the scan also needs to become searchable, run OCR on the sharp original first and compress after.

Hit an exact size when a form names one

Email providers cap attachments around 20 to 25 MB, and the fix is the email preset. Portals are stricter: job applications, exam registrations and government forms routinely demand a number like 100 KB or 500 KB, and “smaller” is not good enough when the upload check is a hard gate.

For those, use the exact-size targets instead of guessing with sliders:

• Compress to 50 KB
• Compress to 100 KB
• Compress to 200 KB
• Compress to 500 KB
• Compress to 1 MB

A target searches every setting, from 300 dpi down to 50 dpi, and keeps the best quality that fits under the number, whatever the preset says; the result names the setting it used. A file already under the number comes back untouched. On our test files, 5 phone photos (14.7 MB) land at 32 KB on the 50 KB target, and a 10-page colour scan reaches 637 KB on the 1 MB target with its pictures at 72 dpi. The same scan cannot get under 100 KB (198 KB at the strongest setting), and the tool says so rather than pretend.

One trap worth knowing: portals do not agree on what a KB is. Some count 100 KB as 100,000 bytes, some as 102,400. The targets here count a KB as 1,000 bytes, the stricter reading, so a file that passes our check passes either way. No guessing, no bounced upload, no third attempt.

Check the result before you send it

Compression you cannot verify is a gamble, so the result screen shows its work:

• Before and after sizes, rounded, counting 1 KB as 1,024 bytes the way Windows does. With a size target there is also a line in the target’s own units (1 KB = 1,000 bytes), such as “Under 100 KB (58 KB)”, so the comparison with the limit is exact.
• A per-page breakdown for the 5 pages that changed most, so you can see where the saving came from and open exactly those pages for a quality check.
• No silent failures. A result that is not smaller is never delivered. If re-encoding gained nothing, you get your original back with a sentence saying why, not a “compressed” file of the same size.

Then do the 30-second human check: open the result, zoom to the smallest text (it is untouched, so it will be sharp), and glance at the photos or scanned areas at reading size. If a picture matters at high zoom, rerun with the print preset or a looser target.

And keep your original. Re-encoding is one-way: no tool can reinvent detail that was discarded. The habit that never backfires is 2 copies: the full-quality file you keep, and the travel copy you send.

What compression cannot do

This is the part most compressor pages skip, so here it is plainly.

A text-only file barely shrinks. A PDF that is mostly text is already a few KB per page; there are no images to re-encode, and the remaining housekeeping savings are small. The size is honest. Do not run it through 3 tools hoping for a different answer.

An already-optimised file barely shrinks either. If the images were already resampled and re-encoded (by a previous compression pass, or by an export that did it well), a second pass finds little to save and would only stack JPEG damage. The result screen will tell you, and hand back the original rather than pretend.

Compression does not change what the document is. It will not remove pages, crop margins or turn 60 pages into 10. When only part of the file matters to the receiver, extracting those pages beats any compressor, and when a hard cap cannot be met at readable quality, splitting into parts is the honest route:

Split PDFBy ranges, every N pages, by bookmarks or size. Free, runs on your device.Open

Sometimes the fix belongs at the source. A huge PDF built from phone photos is better rebuilt with smaller images in the first place (the pictures to PDF guide covers that side), and a bloated export from a word processor is better re-exported than squeezed; the making a PDF guide shows the settings that matter. Compression is the fix for the file you have; the source is the fix for every file after it.

If nobody set a cap and the file sends fine, you can also do the rarest thing in this category: nothing. Compression is a fix, not a ritual.

On a phone, it works the same

Because the compressor is a web page, not an app, it runs wherever a browser runs: Safari on an iPhone, Chrome on Android, Edge on a work laptop with no install rights. Pick the file from Files or Downloads, compress, save the smaller copy back, and share it onward to mail, a chat or a portal upload without touching a computer. No app store, no “free” compressor app with ads, no sign-up.

The one honest caveat on a phone is memory. The compressor holds the whole file in the browser’s memory while it works, so a very large file can exceed what an older phone’s browser has to work with. When that happens the tool says so in plain numbers (how much the file needs, and about how much this browser has) rather than failing silently, and the way out is the same as above: split the file first, or use a computer for that one job. There are step-by-step guides for compressing on iPhone and on Android with the phone-specific corners covered.

Nothing is uploaded, and you can check

Every tool on this site runs on your device, in your browser. Your PDF is opened, re-encoded and saved locally; the only things fetched from our server are the page and the engine that does the work. That is not a promise to take on trust, it is checkable in a minute. The engine (4.6 MB) is fetched the first time you drop a file and kept by your browser from then on, so compress one file, turn Wi-Fi off, and compress another. It still works, because there is nothing to send and nowhere to send it.

That matters most exactly when files are oversized: contracts, medical records, scans of IDs for a visa portal, the paperwork Jane Cooper would rather not hand to a random “free online compressor” that quietly processes it on someone else’s server. Here there is no account, no task limit, no watermark, and no server with your file on it. The copy that goes to the portal in Vienna is made on your machine and never exists anywhere else.