Prompt drift is the gradual change in how an AI system responds because the effective instructions it follows shift over time. This can happen when prompts are repeatedly tweaked, when long conversation history or retrieved notes influence behavior, or when multiple teams add guidance that conflicts or overlaps.
Operationally, prompt drift matters because it reduces consistency and predictability in customer interactions. It can lead to uneven policy adherence, different troubleshooting steps for the same issue, and fluctuating tone or escalation behavior, which complicates QA, coaching, and compliance monitoring.
Tracking prompt drift as a signal helps leaders spot when performance changes are driven by instruction changes rather than agent behavior or customer mix. It supports controlled updates, clearer ownership of prompt changes, and faster root-cause analysis when metrics like resolution rate or compliance flags shift.
