GPT 4o returns the input value as output

API
1

I have a list of transactions and I am trying to generate descriptions for each transaction. I am using the following prompt for doing so.
You are an AI assistant who is an expert in internet search and helps people find useful information.
Generate the description for the TRANSACTION given below.
TRANSACTION: {TRANSACTION NAME/PRODUCT NAME}

 Steps to generate description.
    
    Definition/Scope-What goods or services are involved in the transaction? (e.g battery , hard drive , uniform ,shoes, jersey). What is the definition of the product involved in this transaction? 
    Key Attributes- Highlight any key information or key features about good or service. (e.g power ratings , composition if any edible items, size ,shape)   
Typical Use cases-  Which industry or sector does the transaction pertain to (e.g., retail, healthcare, technology)? Explain the typical scenarios where this good or service can be used.
 Keywords/Phrases- Highlight and explain if there are any typical keywords, phrases or general abbreviation present in description while considering the domain of transaction
  	 Example Products- List some similar goods and services as to the transaction.
Common Formats- List any format mentioned in the transaction in which the product or service is sold.

But in some cases GPT 4o returns just the transaction name/product name which I am giving it as input instead of giving me an elaborated description.

Does anyone have any knowledge as to in which scenarios this happens. I have set the temperature to 0 but even in this case it shouldn’t return just the input back.

2

First - no API model is “an expert on internet search” and actually being able to do it. This is not any service offered by API models or endpoints.

For clearer reading of the response elements you want to get:

Steps to generate description.
    
    Definition/Scope-What goods or services are involved in the transaction? (e.g battery , hard drive , uniform ,shoes, jersey). What is the definition of the product involved in this transaction? 
    Key Attributes- Highlight any key information or key features about good or service. (e.g power ratings , composition if any edible items, size ,shape)   
Typical Use cases-  Which industry or sector does the transaction pertain to (e.g., retail, healthcare, technology)? Explain the typical scenarios where this good or service can be used.
 Keywords/Phrases- Highlight and explain if there are any typical keywords, phrases or general abbreviation present in description while considering the domain of transaction
  	 Example Products- List some similar goods and services as to the transaction.
Common Formats- List any format mentioned in the transaction in which the product or service is sold.

“steps” leaves ambiguity, such as whether the AI should stop and wait for more input after completion of one “step”.

I would make it more deliberate:

As response, you only produce a JSON output that strictly follows this schema:
{
    "type": "object",
    "properties": {
      "definition_scope": {
        "type": "string",
        "description": "What goods or services are involved in the transaction and the definition of the product involved."
      },
      "key_attributes": {
        "type": "string",
        "description": "Highlight any key information or key features about the good or service."
      },
      "typical_use_cases": {
        "type": "string",
        "description": "Which industry or sector does the transaction pertain to and typical scenarios for its use."
      },
      "keywords_phrases": {
        "type": "string",
        "description": "Highlight and explain any typical keywords, phrases or abbreviations present in the transaction."
      },
      "example_products": {
        "type": "array",
        "description": "List some similar goods and services related to the transaction.",
        "items": {
          "type": "string"
        }
      },
      "common_formats": {
        "type": "array",
        "description": "List any format mentioned in the transaction in which the product or service is sold.",
        "items": {
          "type": "string"
        }
      }
    },
    "required": [
      "definition_scope",
    ],
    "additionalProperties": false
  }

There are no mistakes in following that JSON, except for those keys that can be optional based on data, you can write out more conditional statements to evaluate if the JSON key should be used based on the input data.

Then “have a list”? The quality will stay high the less list you try to include in one API call.