QR Codes That Actually Scan: Size, Error Correction and Print

QR codes fail for a small number of well-understood reasons, and almost none of them are the fault of the generator. The code either arrives too small for the distance it is being scanned from, too low in contrast, cropped too tight, printed from a low-resolution raster, or carrying so much data that the pattern is too dense to resolve. Each has a straightforward fix.

The useful thing to understand first is that a QR code's difficulty scales with how much you put in it. More data means more modules — the small black and white squares — packed into the same area, and finer modules need either a bigger code or a closer camera. A short URL and a long one with tracking parameters produce visibly different codes, and the long one is meaningfully harder to scan.

That gives you the single most effective lever: put less in the code. A short link is not a cosmetic preference, it is the difference between a code that scans from across a room and one that needs to be held at arm's length.

The practical guideline used across the industry is that a QR code should be about one tenth as wide as the distance it will be scanned from. A code read from half a metre away — a table tent, a business card, a product label — wants to be around 5cm across. One on a wall read from 3 metres wants about 30cm. One on a billboard read from 20 metres needs 2 metres, which is why billboard QR codes so often do not work: nobody budgeted that much space.

This is a floor, not a target. Going larger always helps and costs nothing but layout space, and the rule assumes a reasonably modern phone camera in decent light. Dim lighting, a cheap camera, an awkward angle or a moving scanner all effectively increase the distance, so add margin wherever the conditions are not ideal.

The rule interacts with data length. It assumes a code of moderate density; a code carrying a long URL has finer modules at the same physical width, and needs to be larger than the rule suggests to compensate. Shortening the data is usually easier than enlarging the print.

QR codes carry redundant data so that a damaged or partly obscured code still reads. There are four levels, and each states the proportion of the code that can be lost while still decoding: roughly 7% at level L, 15% at M, 25% at Q and 30% at H.

Higher is not automatically better, and this is where people go wrong. The redundancy is stored inside the same code, so raising the level adds modules — which at a fixed physical size makes every module smaller and the code harder to scan. Choosing H for a clean poster can produce a code that scans worse than the M version of the same content, because you have traded module size for damage tolerance you do not need.

Match the level to the risk. On a screen or clean printed page, nothing is going to be damaged, so M is right and L is acceptable for very long data where every module counts. Reach for Q or H when the code will be on a curved surface, exposed to weather or wear, printed on fabric, or deliberately overlaid with a logo in the centre — which is really just controlled damage.

A QR code needs blank margin around it — the quiet zone — of at least four modules on every side. This is not decorative padding. A scanner locates the code by finding its three corner squares against empty space, and without that space it frequently cannot tell where the code ends and the design begins.

This is one of the most common real-world failures, because the quiet zone is exactly what a designer removes when fitting a code into a tight layout. Cropping to the edge of the pattern, placing the code hard against a coloured block, or setting it on a busy photograph all have the same effect.

Keep the margin, and keep it plain. White or a very light colour is safest. If the code has to sit on a photograph or a dark background, put it inside its own light panel rather than removing the margin.

Scanners look for a strong light-dark difference. Dark modules on a light background is what every decoder expects and what works most reliably. You can tint the modules, but the contrast against the background has to stay high — a mid-grey code on a white background scans poorly even though it looks fine to a person.

Inverting the code, with light modules on a dark background, is the risky one. Many current phone cameras handle it, and enough do not that it should be avoided for anything printed at scale. If a dark design demands it, test on several devices before committing to a print run.

Two more colour traps are worth knowing. A gradient across the modules reduces contrast somewhere in the code by definition, so keep gradients to the background if you use them at all. And a red-on-white code can scan badly under some lighting because of how sensors weight colour channels — dark blue or black is safer.

A QR code is geometry, so it should reach the printer as geometry. An SVG or a PDF scales to any size with perfectly sharp module edges, while a PNG is a fixed grid of pixels that becomes soft as soon as it is enlarged past its natural size — and soft module edges are exactly what makes a decoder fail.

The failure mode here is subtle enough to survive proofing. A 300-pixel PNG blown up to a 20cm poster panel looks acceptable on screen at a glance, prints with visibly fuzzy squares, and scans unreliably in a way nobody catches until the posters are up. If the only option is raster, generate it at print resolution — at 300 dpi a 20cm code is about 2,400 pixels — rather than enlarging a small one.

For anything going to a professional printer, PDF is the most predictable choice, since it carries the geometry and the intended physical size together.

A QR code can hold more than a URL, and using the right encoding turns a two-step interaction into one. A wifi code joins a network directly rather than making a guest type a password. A contact code adds a full vCard to the address book in one action. There are standard encodings for phone numbers, email, SMS, geographic coordinates and calendar events, all of which phone cameras recognise and offer the matching action for.

The one to be careful with is length. A full vCard with address, several phone numbers and a website is a lot of data, and it produces a dense code that needs to be physically larger. If a contact code has to be small, include name, one phone number and an email, and leave the rest to the website.

For URLs, shorten before encoding. Stripping tracking parameters and using a short domain reduces the module count noticeably, which buys either a smaller printed code or a longer scanning distance from the same one.