Prompt of Scan meal label

1. Send Prompt to ChatGPT

• Provide clear, standardized instructions covering all label tags (WF, WMOF, WSSI, WSSF, LI, RI) and the expected output format.
• Utilize ChatGPT-4.1 model with the "pro" search feature enabled (3x more sources, increased limits) for consistent and comprehensive outputs. Confirm this configuration before running any tests.

2. ChatGPT Confirms Understanding

• The model replies to confirm comprehension of requirements and tagging workflow.

3. Submit Data for Review

• You deliver the system’s JSON output (including dish name, ingredients, quantities, nutrition facts, etc.) along with the corresponding food image.

4.Human Visual Check

• You personally (or a reviewer) visually verify the ingredients and main components in the provided photo.
• Searching trusted source as mentioned in Output.

5. Model Comparison

• Results from the human visual check are compared against both the AI output and ChatGPT’s tag-based analysis.

6. Take Notes

• Specific, referenced notes are recorded per label/tag, following the required structure.

Reference Guidelines for Tag Notes

For every note created for a tag, the reference field must clarify the method or trusted source used:

• Visual estimation from the image:

  reference: Visual estimation from the provided image

• Standard nutrition data comparison:

  reference: USDA FoodData Central, EatThisMuch, or comparable nutrition database

• Standard portion size comparison:

  reference: Portion size guidelines from reputable sources