Order from a restaurant menu photo with ChatGPT
Customers photograph a menu, have ChatGPT compose their order, and show the phone screen to restaurant staff. Task: In the Reddit thread, a server says one customer looked at the restaurant menu on his phone, consulted ChatGPT, and then ordered the item it said was best. In the comments, another server describes the fuller pattern: guests photograph a big menu, ask ChatGPT to build an order, then show the generated order on their; tools: ChatGPT. Human role: Staff must verify every recommended item against the actual menu to correct hallucinations.; limitations: Prerequisites include a smartphone with camera and ChatGPT access. Key constraint is model reliability, as it may invent menu items, discounts, or restaurants. Staff approval is mandatory before order submission..
Tools
Published on Agent Work Atlas:
Overview
Customers photograph a menu, have ChatGPT compose their order, and show the phone screen to restaurant staff.
Workflow
In the Reddit thread, a server says one customer looked at the restaurant menu on his phone, consulted ChatGPT, and then ordered the item it said was best. In the comments, another server describes the fuller pattern: guests photograph a big menu, ask ChatGPT to build an order, then show the generated order on their
Reported result
Actual outcomes are unknown due to source access errors. Proposed acceptance check: Measure time saved during ordering versus the frequency of staff corrections required for non-existent items.
Human role
Staff must verify every recommended item against the actual menu to correct hallucinations.
Limitations
Prerequisites include a smartphone with camera and ChatGPT access. Key constraint is model reliability, as it may invent menu items, discounts, or restaurants. Staff approval is mandatory before order submission.
Try it
I want to test a menu ordering workflow. I will provide a photo of a restaurant menu [replaceable input: menu_image.jpg] and my dietary preferences [replaceable input: preferences.txt]. Please use ChatGPT to analyze the image and suggest 3 dishes that match my preferences. Output the suggestions in a JSON format with fields: 'dish_name', 'reason', and 'confidence_score'. Before finalizing, cross-reference each 'dish_name' against the text visible in the menu image to ensure it exists. If a dish is not found, mark it as 'hallucinated'. Do not place any orders or spend money. First, perform a dry run on a sample menu [replaceable input: sample_menu.jpg] and show me the results for confirmation. If the accuracy is below 90%, stop and ask for clarification on the menu text extraction method.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.