A 40 MB PDF is almost never 40 MB of text. It is 40 MB of scanned pages that were captured at 300 dots per inch and stored as JPEG, sometimes with an OCR text layer bolted on top that adds a little more. Understanding that is what makes compression predictable: the lever is the resolution and quality of the embedded images, everything else is rounding error. That is also why there is no honest answer to the question every page title asks. If the document is mostly vector text with embedded fonts, it was never big and there is nothing to win. If it is photographs, the file cannot get small without throwing away pixels. This page links out to a hosted compressor; the notes below cover what the levels actually trade away, when recompressing backfires, and what silently breaks in the output.
What actually makes a PDF large
Look at the biggest objects in the file and the answer is almost always images. A single A4 page scanned at 300 dpi in colour is roughly 1 to 5 MB depending on how much detail the page had, so a twenty-page report can be 40 MB while containing about 400 words. Typewritten text stored as vector drawing plus a subset font is measured in kilobytes, and such a PDF will run through a compressor unchanged. The other contributors are worth knowing about: fonts embedded in full rather than subset, which multiplies the size of a text-heavy document; duplicated objects that a sloppy producer copied instead of referenced; and, in the case of a long document assembled from scans, a layer of metadata per page. A compressor that is honest about its work will mostly re-encode images and drop unused objects; anything that claims to shrink a vector-only PDF by 90 percent is measuring something other than the file.
Lossy levels, and when more compression means a bigger file
Most hosted compressors offer a few presets, from a lossless pass that mainly repacks the file to an aggressive level that downsamples images hard. The difference between them is a visual one: text stays readable for a long way down, because text is high contrast and survives resampling, but a photograph or a halftone scan starts to look soft, then to show ringing around letters. Two practical points. First, a lossless pass on an already-optimised file can return a slightly larger file, because repacking optimises for a different reader and there is nothing left to remove; compare the two results and keep the smaller one. Second, several of these services offer to run OCR during compression, which adds a selectable text layer over a scan. That is genuinely useful for search and for accessibility, and it is also the step most likely to increase the file size, so treat it as a choice rather than a free extra.
What breaks when a PDF is recompressed
Recompression rewrites the file, and a few things depend on the exact bytes. Digital signatures stop validating, because a signature covers a byte range and every byte around the images moves. Tagged-PDF accessibility structure can be flattened, which matters if the document has to stay accessible. PDF/A conformance, which insists on embedded and correctly subset fonts and specific colour handling, can be lost in ways that are invisible until a validator complains. Form fields are frequently flattened to their current values. Bookmarks and page order usually survive, comments often do not. The rule that follows is simple: if a document is signed, is subject to an archival standard, or is going to be filed as evidence, compress a copy and keep the original untouched. And if you are starting from a scan, compressing the source images before building the PDF often reaches the same target size with a better-looking result.
Frequently asked questions
How much smaller will my PDF get?
It depends entirely on what is inside. A scanned document can usually lose 60 to 90 percent at a moderate level, because the images dominate. A text-only PDF generated from a word processor has nothing to remove and may come out the same size or slightly larger after repacking. Compress a copy, compare the two file sizes, and keep whichever is smaller.
Will compression make the text blurry?
Text survives resampling much better than photographs, which is why moderate levels usually stay perfectly readable. The visible damage appears in images first, then as faint ringing around letters at the most aggressive settings. A good rule: check a page with a table and a page with a photo before choosing a level.
Is the file uploaded, and is the service free?
The service linked here processes your file on its servers, so their privacy policy applies - do not send anything confidential through a hosted compressor. The free tier is usable for ordinary documents and imposes per-file and per-day limits that the provider changes over time. For confidential files, Ghostscript or qpdf on your own machine does the same job with no upload.
Does compressing invalidate a digital signature?
Yes. A signature covers an exact byte range of the file, and recompression changes those bytes, so the signature will no longer validate. Never compress the file you signed; compress a copy for sending and keep the signed original exactly as it was.