Why your PDF is 18 MB, and how to get it under the portal's limit

An 18 MB PDF is almost always a scan. Eight pages photographed by a phone app in full colour at 300 DPI, each page sitting in the file as a 2 MB JPEG — or a 6 MB PNG, which is worse. The words on those pages contribute nothing to the size; the pictures of the words contribute all of it. That single distinction decides whether compression will save you or waste your afternoon, so it is worth understanding before you start clicking.

What a PDF actually stores

A PDF is a container of numbered objects. Text is stored as characters plus a reference to a font program. Images are stored as compressed byte streams. Lines and shapes are stored as drawing instructions.

The size difference between those three is enormous:

• A page of ordinary text: roughly 3–6 KB.
• A subset-embedded font (only the glyphs used): 15–40 KB, once for the whole document.
• A fully embedded font: 200–400 KB, and a Devanagari font can be more.
• One scanned A4 page at 300 DPI in colour: 2480 x 3508 pixels, 8.7 megapixels, 26 MB before compression. JPEG at quality 85 brings it to about 1.2–1.8 MB. PNG brings it to 5–8 MB, because PNG's filters are built for flat graphics and do nothing useful for the sensor noise and paper texture in a scan.

Do the arithmetic on your own file. 18 MB across eight pages is 2.2 MB a page, which matches a colour JPEG scan at 300 DPI almost exactly. A 42-page report that is 1.1 MB is 26 KB a page, which can only be text.

The four things that make a file heavy

Scan resolution and colour depth. This is the big one. Going from 300 DPI colour to 200 DPI grayscale cuts the pixel count by 56% and the channels from three to one. The same page drops from 1.5 MB to around 300 KB.

Images saved as PNG inside the PDF. Some scanner apps and most "image to PDF" utilities embed PNG when the source was PNG. A scanned form as PNG is four to six times the size of the identical page as JPEG, with no visible benefit, because a scan is photographic data.

Fonts embedded in full. Word and LaTeX usually subset. Some export paths do not, and a document using four weights of a large family can carry 1.5 MB of font data on a two-page letter.

Duplicated resources. The same header logo embedded once per page instead of once per document. Old revisions kept alive by incremental saves, so a file you edited twenty times still holds twenty versions of the pages. Embedded colour profiles, page thumbnails, and preview images that no one ever looks at. This is the category you cannot see and cannot easily fix, and it is why a two-page invoice sometimes arrives at 3 MB.

Why phone scanner apps inflate files

Open the settings of whichever scanner app you use and you will find three defaults working against you: colour instead of grayscale, "high" or "HD" quality instead of standard, and auto-enhance switched on.

Auto-enhance is the sneaky one. It boosts local contrast to make the page look whiter and the ink look blacker, which adds high-frequency detail everywhere — including in the paper grain. JPEG spends its bits on high-frequency detail, so an enhanced page compresses worse than the raw capture it came from. Add the fact that a hand-held photo covers more area than the page (the tabletop, your fingers, the shadow) and each capture carries pixels you will never read.

The app's own "compress" toggle usually just lowers JPEG quality on the existing pixels. It helps, but it is fixing the symptom.

What re-rendering at a lower quality really does

Compress PDF works by rendering every page to a bitmap, re-encoding that bitmap as JPEG at the quality level you chose, and building a fresh PDF out of the results. That is the honest description, and it has consequences worth knowing before you click.

If the page was a scan, you lose almost nothing. You were already looking at a picture; now it is a slightly softer picture.

If the page was digital text, you lose the text layer completely. Ctrl+F stops finding anything. Nothing can be copied. A screen reader sees a blank page. And the file often gets bigger, because 4 KB of text characters becomes a 200 KB picture of those characters. The tool checks for this and keeps your original if the output would be larger, but the right decision is not to compress that file at all.

JPEG on text has a specific failure mode: ringing. Sharp black-on-white edges get grey fringes and faint ripples beside them. At the Recommended level you will not notice on 11 pt body text. At Extreme, an 8 pt footnote or a fine-print declaration at the bottom of a form can become genuinely hard to read, which matters when a bank officer has to read it.

Ticking Convert to grayscale throws away the colour channels. On a black-ink form, a photocopy or a printed statement, that typically takes another 40–55% off with no loss of information you needed. Do not tick it for anything where colour carries meaning: a seal, a colour-coded chart, a photograph in a property valuation report.

When compression will not help

• Born-digital PDFs. Already compact. Skip it.
• Scans that were already compressed hard. A page saved at JPEG quality 55 has little left to give; you will get 10–15% and visible damage.
• Vector drawings. A CAD plan or a floor layout is drawing instructions, which stay crisp at any zoom. Rendering it to an image can make the file larger and definitely makes it useless for zooming.
• Password-protected files. The tool cannot read the pages until you open the document with the password and save a copy without it.
• Files that are legitimately long. 400 pages at a realistic 120 KB a page is 48 MB and no setting changes that floor. Split it instead.

Before and after, on real documents

Approximate figures from a handful of sample files, measured with the Recommended and Extreme levels. Treat them as a guide to the shape of the saving, not as a promise about your document.

That last row is the one to remember. A text PDF does not belong in a compressor.

Scan it right and you never need this

Six settings, checked once, and a typical eight-page document comes off the scanner at 2–3 MB instead of 18 MB:

1. Grayscale, unless the document has colour that matters.
2. 200 DPI. At 200 DPI an A4 page is 1654 x 2339 pixels, which resolves 8 pt type comfortably. 300 DPI is for OCR of small print and for printing, not for reading on a screen.
3. Document or text mode, not photo mode. Turn auto-enhance off.
4. JPEG output, not PNG.
5. Crop to the page edge before saving. A tight crop removes 15–30% of the pixels in a typical hand-held capture.
6. Flat surface, even light, camera parallel to the page. A sharp capture compresses better than a blurry one at the same setting.

The practical route when you are already stuck at 18 MB

1. Check the portal's real limit and whether it applies per file or to the whole application.
2. Remove pages you do not need. Bank terms-and-conditions pages, blank versos and duplicate scans are free weight: Remove PDF Pages.
3. Run Compress PDF at Recommended and read the size shown beside the result.
4. Over the limit still? Go to Extreme and tick grayscale.
5. Still over, and the portal accepts more than one attachment? Split PDF into parts with ranges like 1-4, 5-8.
6. Still over? Rescan at 200 DPI grayscale. Compression cannot invent quality that a bad capture never had.
7. Before uploading, open the output and read the smallest print on the worst page. If you cannot read it, neither can the person reviewing your file.

Keep the uncompressed original on your device. The compressed copy is for the upload box, not for your records.

Frequently asked questions

Why did my 400 KB PDF come out at 1.5 MB? Because it was a text PDF and compression turned every page of characters into a picture of characters. Text PDFs are already near their floor. Use the original.

The portal says 2 MB but rejects my 1.9 MB file. Check three things: whether the limit is per document or for the whole submission, whether the portal also enforces page dimensions or a maximum page count, and whether it wants a specific PDF version. Some older portals also count the base64-encoded size, which is about 33% larger than the file on disk.

Does compressing break a digitally signed PDF? Yes. Re-rendering rewrites the whole document, so a DSC signature on a GST or MCA filing becomes invalid. Compress first, then get the final file signed.

Is 200 DPI enough if I want to OCR the document later? It is borderline. OCR wants around 20 pixels of letter height, which 200 DPI gives you for 10 pt type and up. For fine print, dot-matrix bills or Devanagari, scan at 300 DPI and keep a separate compressed copy for uploading.