> ## Documentation Index > Fetch the complete documentation index at: https://docs.zenoo.com/llms.txt > Use this file to discover all available pages before exploring further. # How Do I Use AI Research? > Get AI-powered analysis of screening matches with confidence scores, false positive probabilities, and recommended actions. # How Do I Use AI Research? Zenoo's AI research feature analyzes screening alerts using real-time web research and machine learning. It helps you determine whether a screening match is a true positive or a false positive, saving hours of manual investigation. ## What you'll learn * How to trigger AI research on an alert * What the AI insights panel shows * How to interpret confidence scores and false positive probabilities * How auto-triage works * When to rely on AI and when to investigate manually ## How do I run AI research? Navigate to your alert queue and click on a screening alert (PEP Match, Sanctions Hit, or Adverse Media). In the alert detail panel, click the **AI Research** button. The system queues the alert for AI analysis. A spinner indicates processing is in progress. AI research runs asynchronously. You can continue working on other alerts while it processes. You will receive a notification when the analysis is complete. When analysis completes, the AI insights panel populates with the results. You can also access the results by returning to the alert later — they persist on the alert record. ## What does the AI insights panel show? The AI insights panel is divided into several sections: The AI's overall assessment of the screening match: * **Match assessment** — a narrative explanation of whether the matched person in the database is likely the same person as the entity being verified * **Key findings** — bullet points summarizing the most important evidence for or against a true match * **Comparison points** — specific data points compared (name, date of birth, nationality, known addresses) Two numeric scores help you gauge reliability: * **AI Confidence** (0-100%) — how confident the AI is in its assessment. Higher means more certainty. * **False Positive Probability** (0-100%) — the estimated likelihood that this match is a false positive. Higher means more likely to be a false positive. | False Positive Probability | Interpretation | | -------------------------- | ---------------------------------------------------------------------- | | **90-100%** | Almost certainly a false positive — different person with similar name | | **70-89%** | Likely false positive — some similarities but key differences | | **40-69%** | Uncertain — needs manual investigation | | **10-39%** | Likely true positive — significant similarities | | **0-9%** | Almost certainly a true match — investigate thoroughly | The AI suggests a resolution action: * **Approve** — the AI believes this is a false positive and recommends clearing the alert * **Escalate** — the AI is uncertain and recommends senior review * **Investigate** — the AI found concerning signals that warrant deeper investigation * **Decline** — the AI believes this is a true match with high confidence AI recommendations are suggestions, not decisions. You are always responsible for the final determination. Use the AI research as one input alongside your own judgment and any additional evidence. The AI cites its sources: * **Web sources** — links to articles, company registries, government databases that were consulted * **Database matches** — details from the screening provider's database entry * **Entity data** — the submitted entity information used for comparison Click any source to view the original content and verify the AI's interpretation. ## How does auto-triage work? For high-volume teams, Zenoo can automatically triage new screening alerts: 1. When a new alert is created, the auto-triage system checks if it matches the configured criteria (alert type, category, status) 2. If eligible, AI research runs automatically — no manual trigger needed 3. After research completes, the auto-disposition engine evaluates the results 4. If the false positive probability exceeds the configured threshold (e.g., 95%), the alert is automatically resolved as a false positive 5. A percentage of auto-resolved alerts are flagged for **QA sampling** — they appear in your queue with a "QA Sample" badge for spot-checking Auto-triage and auto-disposition are configured by your compliance team's administrator. Not all alert types are eligible, and thresholds vary by category. Sanctions alerts, for example, typically require manual review regardless of the AI's assessment. ## When should I investigate manually? AI research is most effective for: * **Common name matches** — where the AI can compare biographical details to distinguish between individuals * **PEP matches** — where public information about the PEP's role and tenure is available * **Adverse media** — where the AI can assess whether media articles refer to the same individual Investigate manually when: * The AI confidence is below 70% * The alert involves sanctions (even high-confidence AI results on sanctions should be verified) * The entity's risk tier is already High for other reasons * The false positive probability is in the uncertain range (40-69%) ## What's next? Apply AI insights during your alert review workflow. See how AI findings contribute to risk assessments. Share AI research findings with colleagues via comments. AI research actions and auto-dispositions are logged for compliance.