AI transparency
DistrictWise uses AI to answer your policy questions. This page says plainly what that means, what model does the work, and — just as important — what the AI does not do.
What the AI does
When a staff member asks a question, DistrictWise first retrieves the specific passages from your district's own adopted board policies — and, where relevant, the California Education Code — that bear on the question. The AI model then writes an answer from those retrieved passages, with a citation back to each one, a risk indicator, and a suggested next step. This is retrieval-augmented generation: the model doesn't answer from memory or from the open internet — it answers from what we handed it, and every claim it makes is traceable to a specific policy or statute.
Which model, and who processes your questions
DistrictWise uses Claude, made by Anthropic, accessed through Anthropic's standard commercial API. Only the masked version of your question (see below) and the retrieved policy text are sent to the model — never an un-masked identity. Anthropic's standard API terms mean customer inputs and outputs are not used to train Anthropic's models.
What we do with your questions
This is the sentence we hold ourselves to everywhere — in this page, in our privacy policy, and in the Data Processing Agreement every district signs:
"Your questions are never used to train any publicly available AI model, are never used to answer another district's questions, and never leave the education context."
The "never answers another district's question" part isn't a policy promise we're asking you to trust — it's architecturally enforced. Each district's policies live in their own search index; a query from your staff physically cannot touch, or be touched by, another district's data. There is no shared pool anonymized data could leak into, because no cross-district pool exists. Full technical detail: security overview.
What DistrictWise is not
DistrictWise does not provide legal advice, and nothing in the Service creates an attorney-client relationship. The risk indicator and suggested next step on every answer are triage aids — built to help staff decide when a situation needs to go to district leadership or legal counsel, not to replace that judgment. When an answer is flagged for escalation, that is the system working as designed: getting the right situations in front of the right humans, faster than they might otherwise get there.
Humans stay accountable
DistrictWise cites; your staff decide. Nothing the Service outputs takes an action on its own — every answer is a starting point for a human to verify against the cited source and apply their own judgment, informed by the district's own leadership and counsel where needed.
Known limits — stated plainly
- PII masking is strong, not perfect. Automated name detection can miss uncommon names; the interface shows what was caught so a user can correct it, and staff are advised not to paste rosters.
- Answers are only as current as the corpus. Every answer cites the specific policy version it drew from and the date it was last confirmed current.
- Like any AI system, the model can make mistakes. That is exactly why every claim is cited back to source — verify before you act, every time.
Questions about how this works for your district specifically? Ask us directly.