What refreshes when
| Surface | Cadence | What changes |
|---|---|---|
| Daily metrics — mention rate, share of voice, average position, sentiment, citation rate | Daily | Yesterday’s responses across every tracked platform are parsed and aggregated; the latest values appear in your dashboard each morning. |
| Source Intelligence — citations, source breakdowns, priority gaps | Weekly | Citation data is re-aggregated and the SI brief is regenerated each week, so the priority gaps and source-level guidance reflect a full seven days of fresh data, not a single noisy day. |
| Tasks | Weekly | A fresh set of tasks is generated each week from the latest SI brief, so the queue stays anchored to current opportunities rather than carrying stale items forward. |
| AI Narrative + Recommendations | Weekly | The narrative profile, perception-gap analysis, and prioritized recommendations regenerate together, so the strategic context and the actions Cited recommends always move in step. |
| Impact Scores | On every Narrative refresh | Scores are recomputed against the latest pipeline data, so the priority order you see in the dashboard reflects current competitive context, not a stale snapshot. |
| GEO Score | On demand | The site scan runs when you request it (free scan, audit, or re-scan after fixes). It is not on an automatic schedule. |
What “daily” actually means
Each daily run produces one data point per prompt per tracked platform. Dashboard metrics for any given day are computed from the aggregated results of that day’s run, not from any single response. Week-over-week and month-over-month trends are built from the series of daily aggregates. A single day’s movement can still be noisy because LLM responses are inherently non-deterministic. For trend decisions, two-week moving averages are more reliable than single-day snapshots — see data freshness for guidance on how long to wait before reacting.Why not real-time
Three practical reasons rule out real-time monitoring: Cost. Measuring at the prompt-per-minute level across multiple AI platforms would be prohibitively expensive at any meaningful brand scale. Non-determinism. Single-run measurements are noisy. Daily aggregation across multiple runs produces more stable metrics than a continuous stream of single-point readings. Platforms themselves update slowly. The underlying training data and retrieval behavior of AI platforms changes on multi-week cycles. Faster sampling on Cited’s side does not produce faster signal — the bottleneck is the platforms, not the dashboard.Data availability timing
After the overnight pipeline completes, updated dashboard data is available within minutes. Brands checking their dashboard in the morning see the previous day’s data — no multi-day lag between collection and dashboard availability.Related concepts
- Non-determinism — why daily aggregation matters
- Data freshness — how to interpret confidence in the data
- How we generate prompts — what prompts the pipeline runs
- Impact scoring — how recommendation priority is recomputed each cycle
Frequently asked questions
Why don't Tasks and the AI Narrative refresh every day like metrics do?
Why don't Tasks and the AI Narrative refresh every day like metrics do?
Because day-over-day movements in AI visibility are noisy. Regenerating Tasks or the Narrative every day would produce a churning recommendation list that mostly reflects non-determinism, not real change. Weekly refreshes give the signal time to stabilize so the actions you take are based on directional movement, not single-day fluctuation.
What happens if a pipeline run fails?
What happens if a pipeline run fails?
Individual prompt failures are retried automatically. If a full run fails, the previous day’s data persists in your dashboard until the next successful run, and operational alerts fire to the Cited team. You will not see stale-versus-fresh data mixed in the same view.