LLM Output Quality Degrading Over Time in Automated Lead Gen Workflows

Hey Clay community 👋 Rasmus here, I build automated lead gen workflows for B2B clients in the DACH market. New to the Slack, excited to learn from you all! Quick question: LLM output quality degrading over time Seeing an odd pattern, same table, same prompt, but leads added weeks later get worse LLM outputs (formatting issues, style drift). No re-runs, just new rows processed over time. Has anyone run into this? Any tricks to lock in consistent quality for long-running tables? Thinking: stricter output formatting, model pinning, validation steps?