What it does

The Semantic Index is the conceptual map of your organisation's Microsoft Graph content that lets Copilot ground answers in meaning, not just keyword matches. It's the retrieval layer that finds the right content to feed the model.

Key facts

• It builds vectorised representations of users' and the tenant's content so Copilot can find relevant material by concept.
• It's built and maintained automatically once Copilot is provisioned — there's no separate product to buy or run.
• It strictly honours permissions; the index only ever returns content the querying user can access.
• It underpins the grounding step across the Copilot apps, so its quality shapes answer relevance everywhere.

When to use / skip

This isn't a feature you switch on or off — it's plumbing. You care about it when you're explaining why answers are good or bad, and why content hygiene matters. There's nothing to "use" directly.

Configuration decisions

• Nothing to configure in the index itself; your levers are content hygiene, permissions and what's stored in the Graph.
• Whether your data is actually in Microsoft 365 — content sitting in file shares or third-party systems isn't indexed unless connected.

Gotchas

• It takes time to build and update for a tenant; brand-new deployments can feel underwhelming until it's populated.
• It only knows content in the Microsoft Graph. Data outside M365 is invisible to it without connectors.
• "Respects permissions" cuts both ways — it also faithfully surfaces anything that's over-permissioned.

Consultant notes

• Use this to explain grounding to sceptical stakeholders: garbage or oversharing in the Graph directly shapes Copilot's answers.
• Set expectations on newly-provisioned tenants — the index needs time and content before Copilot feels sharp.
• The takeaway for every project: content and permission hygiene isn't housekeeping, it's the substrate Copilot reasons over.

---

Worth revisiting after the next release wave.