How to photograph and scan documents so they read correctly

Most document-reading mistakes do not start in the engine. They start in the photo: a shadow across the total, flash glare on the date, or a page shot from too far away. Here is what actually matters when you photograph or scan a document, and what you can forget about.

Light

• Daylight or even room lighting beats everything else. Strong light from one side casts the phone's shadow on the page.
• No flash on glossy paper or through a plastic sleeve. The reflection wipes out exactly the area in the middle.
• No hand shadow: hold the phone so the light comes from the side or the front, not from behind you.

Angle and framing

• Shoot from above, parallel to the page. An angled shot shrinks the far lines, and that is where small digits disappear.
• The whole page in the frame, with a small margin around it. A line cut off at the bottom is usually the total.
• Fill the frame. A page that takes up a quarter of the picture loses its small details.

A photo taken with the phone held sideways is turned upright using the orientation information the phone stores in the file. A page photographed crooked stays crooked, though, so it pays to shoot it straight in the first place.

Sharpness and resolution

• Tap the screen to focus on the text, and hold the phone steady until the photo is saved.
• On a scanner, 300 DPI is enough for almost any document. Higher resolution makes the file bigger without improving the reading.
• A simple test: zoom in on screen until the smallest digits fill it. If you cannot read them with certainty, the engine will struggle too.

Format

• PDF from a scanner, JPG or HEIC from a phone: all of them work. Files up to 20MB can be uploaded.
• Sending through a messaging app compresses photos. If someone sends you a photographed document, ask for it as a file rather than a photo.
• A screenshot of a screenshot loses quality every time. Upload the original file when you can.

Special documents

• Multi-page documents: scan to a single PDF. Each page is read on its own and the results are kept together.
• Tables: the whole table should be in one image. A table split across two photos comes out as two partial tables.
• Long receipts: one photo of the whole receipt, even if you have to step back. More in the receipts guide.
• ID documents: photograph only what is needed, and delete the photo from your phone when you are done.

Troubleshooting