Many documents, like application forms, delivery checklists and inspection reports, carry key information in checkboxes rather than text. The AI Vision Engine reads your document visually, the way a person would, so it can tell whether a box is checked and return that as data.
In this article, we'll show you how to set up your mailbox and fields to extract checkbox data. Let's see how!
Before you start
Checkbox extraction works with the AI v3 Vision Engine. Text-based engines and OCR templates do not read checkboxes.
If a template matches your document, Parseur uses the template instead of the AI engine. Make sure no template matches the documents you want to process with AI.
Note: AI extraction is probabilistic. Faint, partially filled or hand-drawn marks may not always be detected, so review a few results before you automate your workflow.
Step 1: Select the AI Vision Engine
Open your mailbox and go to Mailbox Settings > General > AI engine.
Select AI Vision Engine v3 if it isn't already selected.
Click Update:
Step 2: Create a field for each checkbox
Go to the Fields page of your mailbox.
Click New Field and give it a clear name.
Keep the Output format set to Text.
A W-9 tax document with sample data
If your document has a list of checkboxes (like a checklist with many items), create a Table Field instead, with one column for the item label and one for its checked status.
Step 3: Add instructions
In the field's Instructions, describe which checkbox to look at and what to return:
Adding instructions on the fields tab
Tip: To make sure you always get the same values, use the Restrict to a list of choices option (for example, Yes and No).
Step 4: Test and review
Upload a sample document, or reprocess an existing one.
Check the extracted values against the document:
Line 3a is properly accounted for in the output
Adjust your instructions if a checkbox is missed or misread.
Tips for better results
Tip #1: Name the checkbox by its label. Refer to the text printed next to the box, and say where it sits on the page if the same label appears more than once.
Tip #2: Say what counts as checked. Mention ticks, crosses, filled boxes or circled options if your documents use them.
Tip #3: Keep instructions short. Two or three sentences per field is enough. Overly detailed instructions can confuse the AI and reduce accuracy.