AuraCite AI Visibility (GEO)

AuraCite AI Visibility (GEO)

AuraCite tracks how AI engines (ChatGPT, Gemini, Perplexity, …) mention, rank, cite, and recommend brands — "Generative Engine Optimization" (GEO). This skill helps you answer visibility questions using the AuraCite MCP connector (server name auracite).

Prerequisites

The auracite MCP server must be connected. It needs an AuraCite API key in the AURACITE_MCP_TOKEN environment variable. Get a read-only key (mcp:read) from the AuraCite app under API Keys (admin-provisioned during the pilot) — that scope is enough for everything in this skill. Check the connection with /mcp — if it isn't listed, see the plugin README.

Tools available with a read-only key

Tool	Use it for
list_brands	List the brands tracked in the project
get_dashboard_stats	Mentions / citations / sentiment summary (period 7d/30d/90d)
get_mentions	Brand mentions detected in AI answers
get_citations	Direct citations/links a brand earned in AI answers
list_prompts	The tracked GEO prompts (the questions AuraCite asks the engines)
list_competitors	Auto-detected competitor brands
get_visibility_score	Composite 0–100 GEO score with letter grade
get_share_of_voice	Share-of-voice % across AI engines
get_trend_data	Time-series for mentions/citations/sentiment
get_ai_engine_breakdown	Per-engine breakdown (ChatGPT/Gemini/Perplexity/…)
compare_brands	Two brands side-by-side

tenant_id and project_id are injected server-side from the API key — you cannot read another tenant's data, and you don't need to pass them. Most tools need a brand_id or project_id you can get from list_brands.

Good working pattern

1. list_brands to find the brand/project the user means.
2. get_dashboard_stats or get_visibility_score for the headline picture.
3. Drill in with get_mentions / get_citations / get_ai_engine_breakdown / compare_brands.
4. When the user asks "what should we do about it," reason from the evidence (which engines, which prompts, which competitors are winning citations) — don't invent numbers the tools didn't return.

Safety / scope

• This skill is read-only. Scan/ingestion, brand/prompt creation, and credit-spending tools are hidden from read-only keys and require an mcp:write/mcp:spend key plus an explicit AuraCite approval flow. Don't tell the user a write happened unless a write tool actually returned success.
• Cost-bearing actions (new scans, LLM/provider calls) consume AuraCite credits — never trigger them speculatively; confirm with the user first and report the actual cost the tool returns.