AI in student support should show its work
A summary that cannot be traced is not useful to a team making decisions about a child.
The EduNexus Team4 min read
The bar is higher here
AI in a student-support context is not a writing convenience. It sits next to decisions with real consequences, so it has to meet a higher bar than plausible prose.
The rules we hold AI to
- Authorized information only. The model sees what the person asking is already permitted to see, and nothing else.
- Facts separated from interpretation. A summary should be explicit about which statements are recorded events and which are inferences.
- No diagnosis. AI does not label students or suggest eligibility determinations.
- Sources attached. Every claim links back to the record it came from, so a team can verify rather than trust.
- Human review. A person confirms output before it informs a decision.
- A way to report inaccuracy. Anyone reading a summary can flag it, and the flag is recorded.
Governance is a platform feature
These rules are not written into each module separately. They live in the shared AI layer: model registry, prompt templates, privacy handling, usage logging, and review status. Administrators can see what was generated, by whom, for what purpose, and whether it was reviewed.
AI should reduce the effort of understanding a student's story. It should never obscure where that story came from.