Two New Copilot Admin Controls: Agent Usage and Authoritative Content

Two controls, one underlying problem

Microsoft's November 2026 Copilot release notes contain two new Microsoft 365 admin center capabilities that look unrelated and are not. One tells you which agents people are actually using. The other tells Copilot Search which content to trust.

Both exist because the same thing happened to most tenants over the last eighteen months: agents and content multiplied faster than anyone governed them.

• Agents usage report. A new report scoped to agents in Microsoft 365 Copilot and Copilot Chat, covering usage by both licensed and unlicensed users. It includes total active agent users, total active agents, and segmentation of active users by licence.
• Authoritative content management for Microsoft 365 Copilot Search. A centralised governance layer at Copilot > Search > Authoritative content, where you specify SharePoint sites deemed authoritative to improve their relevance and ranking in Copilot Search. Administrators with the Global admin or Global reader role can view, add, delete, and bulk import or export up to 100 authoritative sites. Roadmap ID 503553.

Taken together they address visibility and precision: what is being used, and what should win when Copilot has to choose.

The agents usage report closes a real blind spot

Agent sprawl is a governance problem that has been hard to argue about, because until now the people responsible for it could not see it. Makers could build. Admins could set policy. Nobody had a single view of which agents were live and who was touching them.

The report's design has three properties worth understanding before you read it.

It spans both licence states. Usage from unlicensed users is included. That is the detail that makes this report dual-purpose: it is a governance instrument and it is also the only clean input to a licensing decision, since it shows demand from people who do not currently hold a Copilot licence. If you are working through the November fair-use limits on Copilot Chat, this is the report you need — we cover that decision in Copilot Chat fair-use limits.

It counts active agents, not published agents. The gap between those two numbers is the actual finding. A tenant with 180 published agents and 11 active ones does not have an adoption problem; it has an inventory problem, and the remediation is retirement, not promotion.

Segmentation by licence is where the anomalies surface. An agent with heavy unlicensed usage and no licensed usage is usually either a genuinely broad internal utility or something that escaped its intended audience. Both are worth a look, for opposite reasons.

What the report does not do is tell you whether an agent is well-built, permissioned correctly, or grounded in appropriate data. It is a usage lens, not a quality or security audit. Pair it with activity auditing — see auditing Copilot activity with Purview — rather than treating a usage report as evidence of control.

Authoritative content is a relevance control, not a permission control

This is the one most likely to be misread, so it is worth being precise.

Marking a SharePoint site as authoritative enhances its relevance and ranking in Microsoft 365 Copilot Search. It does not restrict access, it does not hide anything, and it does not stop Copilot from reaching other content. Nobody loses access to anything because you did or did not mark a site.

What it fixes is a quality problem that shows up as a trust problem. In a large tenant, the same policy, template or price list typically exists in several places: the real one, a departmental copy, and two drafts from a migration nobody finished. Copilot Search has no way to know which is canonical, so answers get grounded in whichever document looks most relevant. Users notice that Copilot cited the wrong version, and they stop trusting it — which is a far more expensive outcome than the stale file itself.

The practical limit shapes the strategy. One hundred sites is generous for canonical sources and nowhere near enough to list everything that matters. So this is explicitly a curation exercise:

• Start with the content that produces embarrassing answers when it is wrong. HR policy, security standards, pricing, regulated procedures, brand and legal templates.
• Pick the system of record, not the busiest site. Traffic is not authority. The site everyone visits is often the one with the convenient copy.
• Use bulk import and keep the list in version control. It is exportable, which means it should live beside your other governance configuration and be reviewed on a schedule, not edited ad hoc and forgotten.
• Do not spend the budget on breadth. Ninety-nine mediocre entries dilute the signal you are trying to send.

If your Copilot Search answers are poor for reasons that are not about ranking — missing content, unindexed repositories, connectors that were never set up — authoritative content will not help. That is a grounding problem, covered in grounding Copilot with SharePoint and Graph connectors.

Who does this work

Both controls sit with the platform owner, and the authoritative content list explicitly requires Global admin or Global reader. That is a high privilege bar for what is, in substance, a content curation decision — and the people who know which site is canonical are usually not the people holding those roles.

That mismatch is the thing to design around. The admin holds the control; the business owns the answer. If you do not have a standing forum where content owners nominate canonical sources and a platform owner applies them, the list will be populated once by whoever had access and then never revisited. Organisations that already run a Copilot centre of excellence have the right venue for this — see building a Copilot center of excellence. Those that do not should decide who arbitrates before they start adding sites, because the first real argument will be about whose version wins.

A first-week sequence

1. Read the agents usage report before changing anything. Record active agents against published agents and keep the number. It is your retirement baseline.
2. Look at the unlicensed segmentation separately. It answers a licensing question, not a governance one, and mixing the two produces bad decisions in both directions.
3. Draft the authoritative list offline, with content owners. Target the sources whose wrong answers would cause real damage. Aim well under the hundred-site limit.
4. Bulk import, then export the list and commit it alongside your other Copilot configuration.
5. Set a review cadence. Both of these degrade silently — agents accumulate, canonical sources move during reorganisations and migrations.
6. Re-read the usage report after a retirement pass to confirm the agents you removed were genuinely idle.

What these do not solve

Neither control is a security boundary. The usage report shows activity, not appropriateness. Authoritative content shapes ranking, not access. If Copilot is surfacing content a user should not see, that is a permissions problem and it was a permissions problem before Copilot existed — Copilot simply made it discoverable.

Worth being clear-eyed about scope: these are good, overdue administrative controls that make two previously invisible problems visible. Making a problem visible is not the same as fixing it. The value arrives when somebody acts on what the report says.

Sources

• Microsoft 365 Copilot release notes - agents usage report; authoritative content management for Microsoft 365 Copilot Search (Roadmap ID 503553)
• Partner Center announcements, November 2026

Feature availability rolls out progressively by tenant. Confirm in your own Microsoft 365 admin center before planning around a specific control.