What Kevin can do

Kevin works by calling a fixed set of capabilities against your site’s stored data. Nothing he does triggers a new query to an AI platform.

Read

Capability	What it answers
Metrics	Prompt Visibility Rate, Share of Voice and Sentiment, overall and per platform, matching the Dashboard’s latest full run
Changes	What moved between runs: prompt movers, platform shifts, competitor swings
Competitors	The leaderboard and where each rival beats you
Prompts	Find a prompt by wording, read its answers and its per-platform detail
Citations	Which domains and pages AI cites for your prompts and whether you are among them
How AI describes you	The words AI reaches for and the constraints it raises
Site health	GEO Audit score, Agent Experience score, what is blocked or missing
AI traffic	Visits AI platforms sent you, by platform and landing page (needs Google Analytics)
Actions	Search the Action Centre by keyword, area (on-site, off-site, competitor) or platform
Content opportunities	What is worth writing, ranked
Pages worth improving	Which existing pages to rework, with the effort band, and exactly what to change on one of them
Content performance	Whether published pieces are being cited yet

Capability	What it does	Who can use it
Add a card to the board	Creates an Action Centre board card from the conversation	Any member
Update a card	Moves it between columns, adds a comment, changes labels	Any member
Start a content draft	Opens a Content Studio run for a prompt you name	Any member
Report a glitch	Sends a bug report with the page and context attached	Everyone

Metrics are written out in full: Prompt Visibility Rate, never an abbreviation.
A high figure on a branded prompt is never reported as a win. See Branded prompts.
Stale data is flagged before it is quoted. He never guesses when the next run will finish.
He never quotes the retired composite score. There is one headline number.
Only pages that exist in the sidebar are pointed to. Internal analyses that feed the Action Centre are described plainly, not as places to go.

If you would rather ask from Claude, Codex or Cursor, the MCP server exposes the same read capabilities as tools.