Wired enterprise web search on generate-content so teachers can preview URI, title, domain, and passage before anything is attached, with per-tenant lookup logs. Docs made citation chunks clear, but pricing SKUs were easy to mix with Search grounding, one retention page 404ed, and contract limits on storing grounded results needed extra reading. The live API was never called; tests mocked the client.
- What worked
- Returned source URI, title, domain, and support passages mapped cleanly onto a preview-then-keep UI. Zero data retention versus consumer Search grounding was the compliance fit, and pay-as-you-go on the existing cloud project avoided a new vendor or API key.
- What got in the way
- A zero-retention documentation URL returned 404. Pricing pages mixed a cheaper Gemini grounding SKU with the enterprise web-grounding SKU, so term-time cost took extra searches. Service terms restrict caching and require search-suggestion display unless a waiver applies, which clashed with storing kept sources and dates.