How to compress a PDF, and why some will not shrink

Most advice about compressing PDFs skips the only question that matters: is there anything in your file to take off? Answer that first and you will either save a lot of time or stop trying to solve a problem that does not exist.

What makes a PDF big

A PDF page is a set of drawing instructions: put this text in this font here, fill this shape, draw this image into this box. Instructions are tiny. A forty page contract with its fonts embedded is perhaps a megabyte, and no compressor will meaningfully improve on that.

What makes a document enormous is images, and they arrive in three ways.

A scan. Every page is a photograph of a piece of paper. Twenty pages at 300 dots per inch is twenty pictures of roughly 2,480 by 3,508 pixels, and if the scanner stored them losslessly that is easily 100 MB.

Photos dropped in. A phone photo is twelve megapixels. Ten of them in a report is a large file however it is stored.

Exports from older software. Some programs store placed images uncompressed, or as lossless data when a JPEG would have been invisible from it. These are the files that drop by ninety percent with no visible change at all, and they are more common than you would think.

There is a fourth, smaller cause worth knowing: embedded fonts. Each family costs a few hundred kilobytes. That is why a two page letter can be 5 MB, and it is also why no image compressor will help that particular file. What a PDF actually is has more on what is in there.

Work out which you have, in ten seconds

Open the document and try to select a line of text.

If the cursor draws a text selection, it is a born-digital document. Its text costs nothing. Whatever size it is, the size is in its pictures or its fonts, and if it has few pictures, it will not shrink.

If the cursor draws a rectangle, you have a scan: every page is a picture. This is the case where compression works dramatically, and also the case where it is most visible, because the text you are reading is part of the picture being compressed.

The number that decides everything

Compression tools ask for a quality setting, a percentage, or a preset with a name like “ebook”. Underneath, almost all of them are doing one thing: reducing the resolution of the images and re-compressing them.

Resolution is the lever, and it is measured in dots per inch, on the page. That last part is what most tools get wrong and it matters more than the quality slider. A 4,000 pixel picture is 480 dots per inch when it is spread across an A4 page and 4,000 dots per inch when it is a stamp in the corner. Only the first is worth reducing; halving the second makes a crisp logo fuzzy for almost no saving.

A tool that reduces every image to the same pixel size cannot tell those apart. A tool that works out where each picture sits on the page can, and leaves the small ones alone. What dpi actually means has the arithmetic if the numbers feel abstract.

Pick the target by what will happen to the document:

• 300 dpi if it will be printed. This is the standard for paper and a scan compressed to it still prints properly.
• 150 dpi if it will be read on a screen. Sharp on any display, and usually a half to a tenth of the original.
• 96 dpi when a form or a mailbox has given you a hard number to get under. Visibly softer if you zoom, and it is what gets a 40 MB scan attached to a message.

On a Mac: try Preview first

macOS has a compressor built in, and it is free and crude.

1. Open the PDF in Preview.
2. Choose File then Export.
3. Set Quartz Filter to Reduce File Size.
4. Save under a new name.

For a large scan this often works well enough to stop looking. Its weakness is that it has one setting: you cannot ask for 300 dpi rather than 96, so a document you wanted to print may come back unprintable. Check the result before you delete the original, and never overwrite the original, because this is not reversible.

On Windows: nothing built in

Windows has no PDF compressor. Printing a PDF back to “Microsoft Print to PDF” is sometimes suggested and is usually a bad idea: it rasterises the page, so your selectable text becomes a picture of text, the file often gets larger, and searching stops working.

So on a PC it is a tool, which means choosing one that does not upload your document.

Why this one is worth being fussy about

The PDFs people need to shrink are, by definition, the ones an upload form has just refused: mortgage applications, visa packs, insurance claims, medical letters, scans of passports and bank statements. That is the most sensitive category of file most people own, and the usual free services take a copy of all of it onto a machine you cannot see.

There is no technical reason for that. The work is not heavy and your own device can do it. Is it safe to upload photos online sets out how to judge a service, and the reasoning applies with more force to documents than to holiday photos.

What a good compressor refuses to touch

Some images should be left alone, and a tool that hammers everything equally will damage a document to save nothing.

Black and white scans. A fax-encoded or JBIG2 page is one bit per pixel and already far smaller than any photographic encoding could make it. Re-encoding it as a JPEG makes it bigger and grey.

Stencil masks. These define an edge. Compressing one as a photograph blurs the exact thing it exists to define.

Unusual colour spaces. Indexed and separation colours need care; a wrong decode produces a document with the colours inverted, which is worse than a large file.

If your compressor tells you how many images it changed and how many it left alone, you can see whether it is being careful. If it only shows a percentage, you cannot.

Stopping it happening in the first place

Most of the largest PDFs people own were made larger than they needed to be at the moment of creation, and that is far easier to fix than compressing afterwards.

Scan at the right resolution. 300 dots per inch is right for text you want to read and keep. 600 is for OCR, very small type or an original you will enlarge. Above that you are mostly recording the texture of the paper, at four times the file size.

Scan in the right mode. A page of black text scanned in colour is three bytes a pixel of slightly-grey paper. The same page in greyscale is a third of that, and in black and white it is a small fraction again. Scanner software defaults to colour because it is the safe choice, not because your bank statement needs it.

Use the scanner in your phone, not the camera. Notes on an iPhone and Drive on Android both square the page up, whiten the paper and compress for text. A photograph of a document is several times larger and harder to read.

Export from Word and PowerPoint with the smaller picture setting. Both offer a standard or minimum size option when saving as PDF, which downsamples placed images on the way out. Choosing it once beats compressing the result afterwards.

When compression is the wrong answer

The file is long, not heavy. A 400 page bundle at a sensible resolution is simply a big document. Splitting it and sending the parts is more useful to the person receiving it than compressing it into mush.

It is text and fonts. Nothing to do. Some tools can subset fonts to only the characters used, which saves a little, and that is the ceiling.

It has to be printed. Below 300 dpi, print quality drops off quickly, and “it looked fine on screen” is the classic way to find that out too late.

It is a scan you need to search. Compress it after running OCR, not before, and keep the original. The OCR layer survives compression; the detail it was read from does not.

The short version

Find out whether your PDF is a scan or a document, because that decides whether there is anything to take off. Choose a resolution by what will happen to the file: 300 to print, 150 to read, 96 to fit a limit. Prefer a tool that measures each picture where it sits rather than shrinking everything equally, keep the original, and be careful where you send a file that a form has just refused, because those are rarely files you would post publicly.

The steps, in short

1. Work out why it is big. Is it a scan, or a document with photos in it? Then there is a lot to take off. Is it text with a logo? Then there is not, and no tool will change that.
2. Choose a target by what happens to it. 300 dpi if it will be printed, 150 if it will be read on a screen, 96 if it has to get under an email limit. Lower is smaller and visibly softer.
3. On a Mac, try Preview's Quartz filter first. File, Export, and set Quartz Filter to Reduce File Size. It is crude and free, and for a big scan it often works.
4. Otherwise use a tool that measures the images. A good compressor reduces each picture to the target resolution as it sits on the page, rather than shrinking everything equally, so small logos stay sharp.
5. Check the result before you send it. Open it and zoom to 100 percent on a page of text and on a photo. If the text is part of a scan, that is where the loss shows first.

Questions

Why will my PDF not get smaller?

Because it is made of text and vector shapes, which are drawing instructions rather than pixels and are already tiny. A compressor works on images; if a document has none, there is nothing to remove. A file of a few hundred kilobytes is usually already as small as it goes.

How do I get a PDF under 5 MB for an upload form?

Reduce the images. Start at 150 dpi, look at the result, and drop to 96 if it is still too big. If it is a scan of many pages and still will not fit, splitting it and sending the parts is often more honest than compressing it until it is unreadable.

Does compressing a PDF make the text blurry?

Not in a normal document: text is instructions and is not touched, so it stays sharp at any zoom. In a scanned document the text is part of a picture, so compression does soften it, and that is the case to be careful with.

Is compressing a PDF reversible?

No. Detail that has been thrown away does not come back. Keep the original and compress a copy, especially for anything you might need to print later.

What is the best dpi to compress to?

300 for anything that will be printed, 150 for reading on a screen, 96 when a limit has to be met. Those are resolutions measured on the page, not multipliers, so they mean the same thing whatever your document's images happen to be.

Why did the file get bigger?

Some tools re-encode images that were already efficiently stored, or store them losslessly. A compressor that hands back something larger has failed, and a good one gives you the original untouched instead.

Can I compress a scanned PDF without losing the text?

The text in a scan is a picture, so it softens with everything else. What you can do is run OCR first, which adds a real text layer, so even if the picture gets softer the words stay searchable and selectable.

Is it safe to compress a PDF on a website?

Most sites upload the document and process it on their own machines. Given that the PDFs people need to shrink are usually applications, statements and scans of identity documents, that is worth avoiding where you can. Tools that work inside your browser do not send the file anywhere.