How does the policy assistant work?
A small Policy Assistant panel appears at the bottom of My Policies — but only when your company has turned on the AI assistant setting (off by default, and independent of the module flag itself).
How it works
- Type a question in plain language ("How many sick days do I get, and do I need a doctor's note?").
- The assistant answers only from the current, published, text-bearing versions of policies that already apply to you — this scoping happens at the database query level (by your company and your own audience), not by asking the AI to behave itself. It cannot see or answer from anything outside that set, including other employees' policies or anything not yet published.
- Every real answer comes with citation chips naming the source policy and version; clicking one opens that policy in the reader.
- If no policy you're covered by addresses the question, or the question is too vague to match anything, the assistant refuses with a plain explanation and points you to HR — this is intentional design, not an error state. It never guesses or free-associates.
- A policy version that has only an uploaded file and no authored text is invisible to the assistant — there's no text for it to read.
The privacy note
The panel discloses this up front: "Answers come only from your company's published policies that apply to you. Questions are processed by an AI provider. This is guidance — the policy text is authoritative." This is a deliberate, documented exception to the platform's usual rule of never sending personal data to an external AI provider — your question text and the retrieved policy excerpts are sent to whichever provider your company has configured. That's exactly why the setting defaults to off and should only be enabled with that trade-off understood.