Documents and AI
Uploading documents and taking photos that read well
Formats, size limit and the four mistakes that make a photo unreadable.
A good share of the documents entering the platform are not PDFs: they are phone photos taken by someone on a building site, in a warehouse, or outside a notary's office. Whether that photo can be read comes down to four things, and all four are fixed at the moment it is taken.
The four things that ruin a photo
| Problem | What happens when read | How to fix it |
|---|---|---|
| Shadow or backlight | Half the page goes dark and cannot be read | Stand with your back to the window, not facing it |
| Skewed document | Confidence drops on almost every field | The platform straightens it, but better to shoot it square |
| Blurry photo | Text comes out doubled and is not recognised | Rest the document on a table rather than holding it up |
| Cropped | Exactly the field you needed is missing | Frame the whole document with a little margin |
Worth knowing
If the document has several pages, one photo per page beats one photo of two pages at once. A shot of a double spread reads badly almost every time.
When the file is too large
Almost everything over 30 MB is a scan at unnecessarily high resolution. Rescanning at 200 dpi solves it without losing anything useful: it is more than enough for text, and the file drops to a fraction of the size. If it is a hundred pages, splitting it by section also makes things easier to find later.
Frequently asked questions
›Which formats are accepted?
For certifying, any: PDF, photo, spreadsheet, video, audio. Signing needs a PDF, and anything else is converted.
›Can I upload several documents at once?
Yes. What is worth avoiding is putting different documents into one file: kept separate they are easier to find and to read.
›Is my original file modified?
No. The original is kept exactly as it arrived. Anything generated — a reading, a signed PDF — sits alongside it.
›Can an external participant photograph from their phone?
Yes, from the link itself, with nothing to install. It is the most common route and the one that works best.
A real case
The situation
A site foreman photographs twelve delivery notes at the end of the day and half come out unreadable.
What you do
- Rests them on the bonnet instead of holding them one-handed
- Stands with his back to the sun
- Takes one photo per note instead of two notes per photo
What you get
All twelve read first time and nobody has to go back to site to retake anything.
The situation
The photo comes out blurred and cannot be read.
What you do
- Retakes it with the document flat and good light
What you get
Reading improves without changing anything in the system.
The situation
The photo is taken with a shadow across the text.
What you do
- Changes position before shooting
What you get
The text comes out legible first time.
The situation
A skewed scan is uploaded.
What you do
- Rotates it before uploading
What you get
Automatic reading works.
The situation
Several pages are photographed in one image.
What you do
- Captures one page per image
What you get
Each page is read separately.
The situation
The document has staples or folds covering text.
What you do
- Flattens it before capturing
What you get
No text is missing from the reading.
This article answers
- how do I upload a document
- photograph a document with my phone
- the file is too large to upload
- which formats can I upload
- the document comes out skewed when photographed