The agent persists a buyer-prompt corpus per workspace so successive runs probe the same questions and you get real time-series data. That’s the right default — but eventually you’ll want fresh angles. Refresh prompts generates a new set deliberately differentiated from your existing corpus.
When you’d use it
- The Run outcome panel says “all gaps already covered” — you’ve acted on (or already have content for) every prompt the existing corpus surfaces. New prompts mean new gaps.
- You changed workspace settings (added a competitor, edited audience profile, updated brand voice). The corpus may no longer reflect the new positioning. The Settings → Prompt-corpus section shows a “Settings updated” badge in this case.
- You’re curious what new buyer questions the agent would surface from a fresh angle.
How to trigger it
Two surfaces, same action:
- Inbox — when the Run outcome panel suggests it, hit the Refresh prompts button on the panel.
- Settings → General → Prompt corpus — same button, also pre-selects Quick scope.
Both open the run modal with a checkbox “Refresh prompts (try fresh angles)” pre-ticked. Costs the same as whatever scope you pick (Quick = 1 credit, Top 50 = 3, Full = 6) — refresh is orthogonal to scope, so you can refresh + probe at any size.
Intent-cluster filter
When refresh is checked, a row of chips unfolds: Comparison, Alternatives, How-to, Pricing, Reviews, Explainer. Pick one or more to bias generation toward those intents — useful when you’re losing on a specific kind of buyer question. Leave none selected for any intent (default).
The agent oversamples 3× when intents are filtered to ensure enough usable prompts make it through filtering.
What stays, what goes
Refreshing doesn’t delete your existing corpus — old prompts stay persisted so non-refresh runs still get time-series data. Only this specific run probes only the new prompts. Future Top 50 / Full runs without refresh will use the original corpus.