Text degrades differently from photographs
JPEG was designed for photographs, where neighbouring pixels usually resemble one another and the eye is forgiving of small errors in fine texture. Text breaks both assumptions. A letterform is a hard black edge against white, which is the highest-contrast transition an image can contain, and it is exactly where JPEG's block-based compression produces its most visible artifacts.
This is why a compressed scan develops faint grey halos and speckling around characters before it shows any other sign of damage — the effect known as ringing. Small print goes first, because the artifacts are a fixed size in pixels and therefore consume proportionally more of a small character. It is also why the same file size looks fine for a photograph and poor for a document.
The practical consequence is that for scans you should protect quality rather than resolution, which is the reverse of the advice for photographs. This tool keeps quality above the point where ringing becomes destructive and reduces dimensions instead once that floor is reached, so text stays clean even when the page gets smaller.
Scan rather than photograph, whenever you can
A phone photograph of a document carries several things a scan does not: uneven lighting across the page, a shadow from the phone or your hand, keystone distortion from not being perfectly square on, visible paper texture, and often a slight focus gradient. Every one of those is detail the encoder must spend bytes reproducing, none of it is information you want, and together they can double or triple the file size needed for the same readability.
If a flatbed scanner is not available, most phones now include a document mode that flattens perspective, crops to the page edge and normalises the lighting to white. The output is much closer to a scan and compresses far better. Failing that, photograph the page under even indirect light, hold the camera square to the page rather than tilted, and fill the frame.
When a PDF is the better answer
If the portal accepts a PDF, a multi-page document is almost always better submitted as one. You avoid the naming and ordering confusion that comes with a folder of separate images, and PDF compression can be applied per page, so a form with one dense page and three sparse ones does not have to treat them identically.
The exception is when the limit is genuinely per-file and tight. In that case splitting a document into separate images, each fitted to the limit individually, gives you more total budget than one PDF constrained to the same figure. Check whether the stated limit applies to each upload or to the submission as a whole before deciding.
Colour, greyscale and what you can safely discard
A great many documents are black ink on white paper, photographed or scanned in full colour for no reason other than that colour is the default. Colour information is genuinely expensive — it roughly doubles what a scan costs to encode — and on a monochrome page it carries nothing but the yellowish cast of the paper and whatever colour temperature the room lighting happened to have.
If the portal does not specifically require colour, converting a document scan to greyscale before compressing frees a substantial part of your budget to spend on the thing that actually matters, which is the sharpness of the text. The exception is any document where colour is evidence: an official stamp, a coloured signature, a security watermark or a photograph embedded in an identity page. If in doubt, keep colour and accept the smaller quality budget, because a rejected document costs far more than a slightly softer one.