● Files never leave your device

Why is my PDF so large?

How to

1. Check MB per page first. Divide file size by page count. Over ~1 MB per page points at images; well under it points at fonts, forms or structure.
2. Open the matching tool. Compress for image-heavy files, Flatten for filled forms — in your browser, nothing is uploaded.
3. Compress, or set a target size. Pick a level, or type the exact size a portal demands and let the tool try a limited set of quality settings; then check the output size and readability.
4. Compare before you commit. The preview shows quality at 100%. For a colour scan of a black-and-white document, try grayscale first.

The short answer depends on how big the file already is

We grouped the documents into three size bands and took the median share of each category. Median rather than mean, because one 6 MB image-heavy outlier skews an average by itself.

File size	Docs	Images	Fonts	Content	Form fields	Structure
Under 250 KB	8	0.0%	11.6%	43.3%	12.8%	19.7%
250 KB – 1 MB	6	0.0%	52.2%	14.9%	2.8%	5.3%
Over 1 MB	8	26.0%	6.7%	29.3%	1.1%	10.3%

Above 1 MB the common wisdom holds: images dominate. Below 1 MB the bytes are somewhere else entirely — in embedded typefaces, in page drawing instructions, and in something almost no guide mentions: the definitions of form fields.

Cause 1 — scan resolution (still the top cause for big files)

Phone and office scanners default to 300–600 DPI in full colour. A single A4 page at 600 DPI colour is roughly a 35-megapixel image. For reading on screen 150 DPI is plenty; for archiving text, grayscale halves it again. This is the one case where the internet's standard advice is exactly right — and our data agrees: the image-dominated documents in our set reached an 85–89% image share.

Document size	Image share
6.6 MB	89.0%
751 KB	85.0%
3.6 MB	70.9%
1.5 MB	60.1%
6.5 MB	31.2%

Fix: run the file through Compress. Scan images are re-encoded at document resolution rather than sensor resolution. A 40 MB scan typically lands at 2–4 MB; if a portal enforces a hard cap, set the target size instead of a quality level.

Cause 2 — embedded fonts, the silent majority

In the 250 KB – 1 MB band the median font share is 52.2%. At the extremes of our set it passes 85%: almost the whole file is not the text you read, but the typeface programs that draw it.

Document size	Font share
664 KB	87.6%
821 KB	85.4%
158 KB	71.2%
359 KB	65.3%
185 KB	53.8%

The cause is missing or partial subsetting. A PDF embeds the typefaces it uses; a good producer embeds only the glyphs that actually appear. A poor producer — or a template that pulls in every weight separately — embeds the full character set once per variant. Five families times four weights is twenty embedded font programs, and none of them shrink further because font files are already compressed.

Cause 3 — form fields, the overhead nobody counts

Every box, checkbox and dropdown in a fillable form lives in the file as a dictionary object: position, type, default value, appearance stream, validation rule. These are not streams, so no compressor touches them — and no guide counts them.

Document	Annotations	Form share
Tax return form	229	26.8%
Information form	124	24.7%
142-page guide	2,958	21.3%
Return schedule	275	21.2%
Short form	85	19.2%

A quarter of a tax form is the definition of the boxes you are meant to fill in. It is tempting to conclude that flattening the form — discarding those definitions once it is filled — recovers a quarter of the file. We measured that too, and it does not.

Fields	Before	After	Change
199	215 KB	327 KB	+52% larger
105	120 KB	170 KB	+42% larger
116	822 KB	874 KB	+6% larger
29	664 KB	670 KB	+1% larger
1	2,360 KB	2,075 KB	−12% smaller

Cause 4 — photos and screenshots pasted at original size

A report with five phone photos embeds five 12-megapixel originals even when each is displayed at postage-stamp size: the page shows the small version, the file carries the large one. Screenshots are worse in a specific way — they are PNG, lossless and large, and on a Retina or 4K display one full-screen capture is a 4000-pixel-wide image.

Compression re-encodes images at the size they are actually displayed, which is why photo-heavy reports and screenshot-heavy decks routinely shrink by 80–90% with no visible difference at reading size.

Which one is your problem? A decision table

Divide the file size by the page count first — that single number narrows it down immediately.

What you have	Dominant cause	What actually helps
Over ~1 MB per page, pages look like photos	Images	Compress; grayscale for B/W scans
A filled-in fillable form	Form fields	Flattening locks it — but does not shrink it (measured below)
Text-only report, 200 KB – 1 MB	Embedded fonts	Re-export; use fewer typefaces
Hundreds of pages of plain text	Content streams	Already near optimal — split instead
Small file, few pages, no images	Structure and form fields	Re-export through a rewriting editor

Every one of these runs inside your browser in Haven PDF. An oversized file never has to be uploaded anywhere to get smaller — which also means a confidential contract stays on your machine while you shrink it.

Compress PDF · Flatten a PDF form · Convert to grayscale

How we measured, and what this does not prove

We walked each document at the object-graph level and assigned every stream to a category by its raw length — the space it genuinely occupies in the file. Image XObjects and streams carrying JPEG, JPX, CCITT or JBIG2 filters count as images; streams referenced through FontFile, FontFile2 or FontFile3 count as fonts; the rest are content. Bytes outside the counted streams were treated as structural overhead, and form and annotation dictionaries were separated out of that total.

Frequently asked questions

My PDF is all text and still huge — why?

Most likely embedded fonts. In our measurements, text documents in the 250 KB–1 MB range had a median font share of 52%, and the worst case was 88%. Image compression cannot touch that; re-exporting through an editor that rewrites the file, or rebuilding it with fewer typefaces, is what helps.

Will compressing make the text blurry?

Real text — typed, not scanned — is untouched by image compression and stays vector-sharp at any zoom. In a scan the text is part of the image, so it is affected; medium levels normally keep it clearly readable.

Why did compression barely change my file?

Almost certainly because there were no images to compress. Eleven of the twenty-two documents we measured contain none at all. Check MB per page: well under 1 MB per page usually means the bytes are in fonts, form fields or structure instead.

Does flattening a form really save that much?

No — and we checked. Field definitions are 19–27% of those files, but flattening nine of them made the median file about 6% BIGGER, with the worst case growing 52%. The definitions go away, yet each field’s appearance then has to be drawn into the page. Flatten to lock a form, not to shrink it.

Is there a safe maximum compression?

Use the preview: compress, look at a dense page at 100%, and only go stronger if it is still comfortably readable. With a target size the tool picks the best quality that fits automatically.

Sources

1. ISO 32000-2:2020 — Document management: Portable Document Format — Normative definition of object structure, stream filters and AcroForm field dictionaries.
2. PDF 1.7 Reference (Adobe; same content as ISO 32000-1) — Font embedding (§9.9), image XObjects (§8.9), annotations (§12.5).
3. Corpus: IRS forms and publications, arXiv papers, public agency reports — All publicly downloadable documents. No personal or private files were used.

Related tools