AI incidents
Datadog-style alerts when AI visibility, mentions, citations, or narratives change between prompt runs.
Overview
AI incidents (/incidents) alert you when something meaningful changes in how AI answers questions about your brand. Obsurfable compares each new prompt run to your previous run and opens incidents for visibility drops, lost mentions, lost citations, narrative shifts, competitor surges, and entity drift.
Each incident includes a root cause analysis with contributing factors and recommended next steps.
Who it's for
- Pro and Agency teams running recurring prompt scans who need to know when visibility changed—not just the latest score
- Operators who want engineering-style observability for AI brand health
Prerequisites
- At least two completed prompt runs on the same company prompts (Pro re-runs or scheduled automations).
- A Pro plan — incidents are generated automatically after each run on paid plans.
Incident types
| Type | What triggers it |
|---|---|
| Visibility drop | AI Visibility Score falls by 10+ points between runs |
| Lost mention | Your brand was mentioned before but not in the latest answer |
| Lost citation | Your domain was cited before but not in the latest answer |
| Narrative shift | The answer text changed substantially for the same question |
| Competitor surge | Citation sources or competitor presence shifted |
| New entrant | A new brand appeared in recommendations |
| Entity drift | Entities associated with the answer changed |
Step-by-step
- Run or schedule prompt runs regularly to build a baseline.
- Open Incidents from the dashboard or go to
/incidents. - Review open incidents — severity is shown as low, medium, high, or critical.
- Read the root cause analysis for contributing factors and suggested actions.
- Acknowledge or Resolve incidents as you investigate.
- Use Compare runs or the linked prompt run for full response diffs.
Email alerts
Enable Email me when AI incidents are detected in Settings. Alerts respect your company's alert_email_enabled preference.
Tips
- Incidents are most useful with scheduled automations — run every ~6 days so changes surface quickly.
- Pair incidents with Entity perception when the incident type is entity drift or narrative shift.
- Resolve incidents after you've acted so your open-incident count reflects current work.
FAQ
Which AI platforms do incidents cover?
Today, incidents are based on ChatGPT with optional web search (Pro) — the same retrieval used in prompt runs. Claude, Gemini, and Perplexity monitoring are coming soon.
Do incidents replace Insights?
Incidents focus on what changed and when. Insights still provide broader AEO/GEO recommendations from a full run. Use both: incidents for alerts, insights for remediation ideas.