Asking Good Questions
Once you’ve configured a provider (see Setting Up an AI Provider), open Ask InsightLens from the sidebar. The chat view has a single text box and a conversation history.
What the assistant can see
The assistant has access to the schema of your local survey database and to the results of queries it runs against that database. It does not see raw PDFs, and it does not see data from any other source. You’re asking questions of the data you imported, nothing more.
Good question patterns
- Be specific about scope. “Which Semester 2 2024 units in Information Systems had a response rate below 20 %?” is much easier to answer than “show me problem units”.
- Name the metric. Overall experience, response rate, comment count, and sentiment are all different things. Pick one.
- Ask for a chart when you want one. “Plot the overall experience trend for ISYS2001 across all semesters” will produce a chart in the conversation.
- Iterate. Follow-up questions build on the previous answer. “Now compare that to ISYS3001” will do the right thing.
Things that confuse the assistant
- Vague superlatives (“which unit is best?”) — best at what? over what period?
- Data you haven’t imported — if you only imported one campus, asking about another will return empty results.
- Non-survey knowledge — the assistant is not a general-purpose chatbot. If you ask it about the weather, it will politely decline.
Verifying answers
Whenever the assistant reports a number, click into the Units page or the Unit Detail page and cross-check. InsightLens is designed so every AI answer is reproducible from the underlying data — if you can’t find the same number by filtering by hand, treat the answer with suspicion and ask the assistant to show its working.
Privacy reminder
When you use a cloud provider, your questions and the database rows needed to answer them are sent over HTTPS to that provider. For sensitive free-text comments, consider running a local model via Ollama instead.