A keyword is a hierarchical label the AI uses to classify what a call was about. Each keyword carries a plain-language prompt that tells the call-analysis layer when to apply it, building a structured tag tree such as "Support -> Billing -> Refund" on top of call audio.
Overview
Keywords power reporting, routing decisions in Smart Assistants, and quality reviews. They form a tree up to four levels deep: a top-level keyword is a broad category, its children are subcategories. Each keyword carries an AI prompt describing the criteria the AI uses to decide whether a call matches.
What it does
After a call ends and is submitted for analysis, the AI evaluates each keyword's prompt against the transcript and tags the call with every keyword that matches, including ancestors. A call matching Refund inherits the Billing and Support tags too. Matching never affects the live caller, only reporting, CDR detail pages, and future Smart Assistant routing.
The Prompt is the most consequential field. Good prompts are specific about what should match and what should not: "The caller mentions a refund, money back, return, or chargeback request" matches cleanly, while "The caller mentions money" is too broad.
See also
- CDRs - where the AI's keyword tags appear on individual calls.
- Smart assistants - the AI IVR feature that can route based on keywords.
- Tags - manual labels applied to users and devices (different from keywords).