Reading your analytics
The analytics view answers one question: what are my players trying to figure out? It turns scattered messages into a short, honest list. Here’s how to read it.
Top questions
The headline view is the questions players ask most, grouped so that ten ways of asking the same thing count as one. Each row shows the question, how often it came up, and how confidently the agent answered it.
Read this as a to-do list for your game, not just your agent. A question near the top that players ask constantly is often a sign the game itself could explain that thing better.
Volume over time
A simple chart of how many questions the agent handled per day. Use it to see the shape of your community’s attention.
- A spike usually lines up with a patch, an event, or a content drop. Players have new things to figure out.
- A steady climb in the first weeks is the agent catching on as more players learn it’s there.
- A sudden drop can mean the agent went quiet. Check the agent isn’t responding.
Topics
Questions are sorted into topics that match your game: careers, economy, a specific world, and so on. The topic breakdown tells you where confusion clusters. A topic that’s large and growing is worth your team’s attention, whether that’s a tooltip in the game or a pinned post in Discord.
Confidence
Every answer carries a confidence signal. The share of low-confidence answers is your early warning system.
- A small, steady share is normal. No agent answers everything.
- A rising share, especially right after a patch, usually means the agent’s data is behind. See how patches get ingested.
- A cluster of low confidence around one topic can mean a spoiler rule is holding answers back, or a real gap worth telling us about.
The questions behind those low-confidence answers get their own view: unmet questions.
Filtering
Narrow any view by date range or channel. Comparing two channels can be useful: your help channel and your general channel often ask very different things.