Verified 2026-05-03 for the agent’s existence, model, and core capabilities. Specific tool count drifts as features ship — verify against the agent registry in
nudg3-workflows/agents/before quoting an exact tool count externally.
Ask NUDG3 Agent
A conversational AI agent accessible from any page in the dashboard. It queries the user’s workspace data, investigates performance issues, manages prompts, and pulls in GA4/GSC data — all through natural language.
Runs on LangGraph Platform with Claude Sonnet 4.6.
What users can do with it
| Ask this | The agent does this |
|---|---|
| ”How’s my visibility this week?” | Pulls visibility overview, compares to last period, highlights changes |
| ”Which providers am I weakest on?” | Analyses per-provider mention rates and identifies gaps |
| ”Compare me to [competitor]“ | Runs competitive landscape analysis across all providers |
| ”What’s driving my low Perplexity score?” | Investigates prompt performance on Perplexity specifically |
| ”Create a prompt for ‘best insurance SA‘“ | Previews the prompt, creates on confirmation |
| ”Retire the worst-performing prompts” | Shows underperformers, retires on confirmation |
| ”What’s my organic traffic trend?” | Pulls GA4 data if connected |
| ”Show me my latest report” | Retrieves the most recent visibility analysis |
Key characteristics
- Data-grounded — every response is backed by tool calls to real workspace data. The agent never fabricates metrics.
- Persistent conversations — 7-day thread TTL. Conversations survive page refreshes, navigation, and multi-day sessions.
- Prompt management — can create and retire tracking prompts with a human-in-the-loop confirmation flow (writes are confirmed, reads are not).
- Integration-aware — if GA4/GSC is connected, the agent uses those tools. If not, it doesn’t try.
AI Search Funnel Phases
The agent structures analysis around three phases (also tagged on every prompt — see Stage 1):
| Phase | Query Type | Example | What it measures |
|---|---|---|---|
| Discovery | Unbranded | ”best CRM software” | Does AI recommend you unprompted? |
| Research | Branded | ”tell me about Acme Corp” | Does AI describe your brand accurately? |
| Purchase | Intent | ”should I buy Acme CRM” | Does AI recommend you to buyers? |
Workspaces without phase classifications are handled gracefully — the agent uses custom content tags for topic-level analysis instead.
Why this matters in sales
The agent is the answer to “do I really have to learn another dashboard?” For non-power users, the answer is no — they can ask the agent in plain English and get a sourced answer. For power users, the dashboard remains primary.
See also
- Integrations — what GA4/GSC unlock for the agent
- Stage 4 — Reports & Boosters — where the agent fetches “show me my latest report” from
- Public API & MCP — programmatic equivalent for AI tools outside the dashboard