A quick diagnosis
Drop the file on Compress PDF and read the three candidates. If Email & web is a small fraction of the original, the file is images and compression will fix it. If all three are close to the original, the file is not images — read on for fonts and leftovers. If Print (lossless) is noticeably smaller than the original, the file has structural waste: duplicated resources, uncompressed object tables, or objects left over from earlier edits. That single screen answers most cases in a few seconds, on your device.
1. Images placed at a resolution nobody will see
A 12-megapixel phone photo is about 4000 pixels wide. Placed on a page and read on screen, 1400 pixels would look identical; on an office printer, 2200. Documents built from photos, product shots and screenshots carry three to ten times the pixels they display. The fix is resampling to the destination’s resolution, which is exactly what the Email & web (about 96 DPI) and Balanced (about 150 DPI) targets do, with the actual output size shown before you commit. Text is never touched, so the document stays sharp and selectable.
2. Scans, which are pictures of every page
A scanned page is a full-page photograph, typically 2500 × 3500 pixels in colour, 300 KB to 2 MB each. A hundred-page scan is a hundred photographs. Two things help. Mixed-raster compression, which Bindery applies to scanned pages on the lossy targets, separates the text into a crisp one-bit mask and stores the background at low resolution — the technique behind the small files good scanner software produces. And a text layer from OCR costs almost nothing while making the document searchable. Rescanning text documents in greyscale at 300 DPI is the best fix of all, when it is still possible.
3. The same image, embedded many times
Slide decks and letterheads do this: the same logo or background image stored once per page rather than once per document. It is invisible in a viewer and obvious in the file size. Compression on any target stores identical streams once; the Print target does it losslessly, which is why a “lossless” pass can still cut a deck in half.
4. Fonts
A document that embeds a few full font families — every glyph of every weight — can carry several megabytes of fonts, which is why a text-only PDF is occasionally large. Most exporters subset fonts to the glyphs used; some do not. Bindery does not strip or re-subset fonts, because removing glyphs can break editing and PDF/A conformance; the honest fix is to re-export from the authoring tool with font subsetting on. Extract images is a useful check here: if it finds few or small images in a large file, fonts or leftovers are the cause.
5. Leftovers from editing
Every time some editors save a PDF they append the changes to the end of the file and leave the old objects in place — “incremental updates”. A form filled ten times, a document annotated over months, or a file that went through several tools can carry several complete copies of its own history. Deleted pages may still be in the file. The Print target rewrites the document from scratch, keeping only what the current pages use; that is the fix, and it is lossless.
What does not help
- Zipping a PDF: images inside are already compressed, so a ZIP saves a few percent at best.
- “Print to PDF” from a viewer: it often rasterises pages and makes them bigger, and it discards the text layer.
- Cropping: the content outside the frame stays in the file. Crop is for the visible area, not for size.
- Compressing a file that is already efficient: Bindery says so rather than degrading it; the 8% rule means nothing is replaced for no gain.
The order that works
- Delete pages nobody needs and extract if only part is wanted.
- For scans, clean and then OCR.
- Compress once, on the target that matches the destination, and read the size.
- Check the result at 200% zoom, especially fine print and signatures.
- Keep the original; the compressed file is a delivery copy.
The compression benchmark publishes measured results on sample files, and PDF too large to email covers the mail-specific caps.