How do I troubleshoot inconsistent AI enrichment results?

Troubleshooting Inconsistent AI Enrichment Results

The issue likely stems from contradictory instructions in your prompts. LLMs may interpret conflicting directives differently, leading to inconsistent outputs.

Root Cause

Your instructions contain conflicting guidance:

• One instruction states: "Return that no relevant information was found"
• Another states: "If no verifiable information is found, return an empty string"

Solution

For Point #4: Change the instruction to:

"I prefer news from the last 6-12 months, including news about product launches, funding, etc. However, if no such news is found, return 'No significant news found.'"

For Section 5: Update the final line to:

"DO NOT fabricate news. If any of the information I asked you to find cannot be found, return a sensible statement such as 'No significant news found'. In any case, DO NOT RETURN an empty string/column."

Key Principle

Ensure your instructions provide one clear directive for handling cases where information cannot be found, rather than multiple conflicting approaches.

Troubleshooting Inconsistent AI Enrichment Results

Root Cause AI enrichment can produce inconsistent results across different rows due to variations in data quality and prompt specificity.

Solution: Iterative Testing Approach

1. Test on a small sample first — Start with 5-10 rows that represent different data types in your dataset
2. Validate prompt reliability — Ensure your prompt works consistently across this mixed sample
3. Scale to full table — Once validated, run the enrichment on your complete dataset

This approach minimizes costs while identifying potential issues early before processing large volumes of data.