Keep the prompt set fixed between runs, because cited domains drift 40.5% to 59.3% in a single month (70-90% January vs July 2025), and changing prompts destroys trend readability.

Why this matters

Profound's ~80,000-prompt-per-platform study measured this volatility; changing prompts confounds your work with background movement.

Implementation

Type: not_automatable

Anti-pattern

Rewriting the prompt set between monthly runs and reading the delta as your impact.