Copilot data sources 2026: Power BI, Dataverse, scanned PDFs

For most of its life, Microsoft 365 Copilot has reasoned over a well-understood set of sources: mail, calendar, Teams, OneDrive and SharePoint, plus whatever a Copilot connector had indexed. Over the past few months that boundary has moved in several directions at once. Power BI semantic models, Dataverse tables, third-party systems reached live over MCP, HubSpot records, scanned PDFs and column metadata in document libraries are all now in scope or arriving. Each is good news for the people who want answers, and a new question for whoever owns data governance in a 50-seat Australian business, because every source Copilot can read is a source it can repeat to anyone with permission to ask.

This note works through the items Frontrow is being asked about, using the Microsoft 365 roadmap and the Copilot release notes as the only sources for what each feature does. For each: what it makes possible, what it needs, and the governance question to settle first.

Power BI integration in Microsoft 365 Copilot

Roadmap item 567891 is rolling out, with general availability listed as August 2026. Microsoft's description: it "enables M365 Copilot to reason over Power BI enterprise data and returns grounded answers directly from Power BI reports and semantic models." Users ask a question in natural language and receive an answer grounded in the model rather than in a screenshot of a dashboard someone pasted into Teams.

Prerequisite: the item sits under Microsoft 365 Copilot, so a Copilot licence is the entry point, and the user needs the Power BI access they would need to open the report themselves. Governance question: which semantic models are fit to be quoted? A curated, certified finance model is a fine thing for Copilot to answer from. A sales model three people built in 2023 and nobody has reconciled since will answer with the same confident tone. Decide which workspaces are authoritative before this rolls out, and confirm row-level security on those models does what the finance team believes it does.

Dataverse data connected to Copilot

Roadmap item 560539 is marked launched, general availability listed as September 2026. Microsoft's description is a single line: "Let people search and query Dataverse business data from Microsoft 365 Copilot." That is the whole published definition, so this note does not elaborate on mechanics the roadmap does not describe.

Prerequisite: a Dataverse environment with data in it, which in practice means Dynamics 365 or Power Apps running on Dataverse tables, plus Copilot licensing on the user side. Governance question: Dataverse holds customer records, case notes and sometimes employee data, and its security roles were designed for model-driven apps, not for a chat interface that can summarise a table in a sentence. Review who holds broad read roles on sensitive tables first, because Copilot will honour those roles exactly as they stand.

Federated Copilot connectors

Roadmap item 501120 reached general availability in April 2026 and was covered in the 23 September release notes. Federated connectors let Copilot retrieve data from third-party sources in real time using the Model Context Protocol (MCP). The release notes make a point that matters for Australian data-handling conversations: these connectors "do not store or index customer data in Microsoft services; access occurs in real time using the user's identity, while administrators retain full governance and control through the Microsoft 365 admin center." They are supported in Researcher, Microsoft 365 Chat and when editing with Copilot in Excel.

Prerequisite: a Microsoft 365 Copilot licence, the connector set up and approved by an administrator in the admin centre, and the user's own identity in the third-party system, since access is in that user's name. Governance question: real-time federation means there is no index to review, which is good for data minimisation and bad for visibility. Frontrow recommends approving federated connectors one source at a time, with the business owner of that source signing off.

Two companion changes help. The 6 October release notes report that admins can now edit query string parameters and user identity mapping rules on an existing connection, under Settings, then Search and intelligence, then Data sources, without a support ticket; the notes describe this as making access scope and content scope self-serve. And roadmap item 502541, launched with general availability listed as September 2026, adds content source filters to improve discovery of connector sources in Copilot, so a user can tell where an answer came from.

HubSpot connector in Researcher

Roadmap item 501292 is launched, general availability listed as November 2025, and was updated on 13 October 2026. The HubSpot connector lets users of the Researcher agent bring HubSpot information into generated responses. Prerequisite: the Researcher agent, which ships with Microsoft 365 Copilot, and the HubSpot connector configured and consented by an administrator. Governance question: CRM notes are where staff write the things they would never put in a report. Before connecting HubSpot, decide whether every Copilot user should be able to ask Researcher for a named client's history, or whether the connector should be scoped to the sales and account teams.

Scanned PDFs and text inside images

Roadmap item 559613 is rolling out, general availability listed as October 2026. Copilot Chat "can now better match search results from Scanned PDFs" in the files a user is working on, content that was previously difficult to search and find. Microsoft notes it works for new or updated working files on a go-forward basis, so the back catalogue does not become searchable overnight; a file has to be touched again.

A related item, 570469, extends the same matching to text inside images embedded in Word and PowerPoint files, with scanned contracts and text in tables given as examples. It is in development, with availability listed as November 2026, so plan for it but do not promise it. Prerequisite for both: a Copilot licence and files saved or updated since the feature reached the tenant. Governance question: scanned PDFs are where businesses keep the documents they were least careful about: signed employment contracts, bank details on supplier forms, identity documents from onboarding. If those sit in a broadly shared library, they are about to become findable by everyone who can open it. This is the item Frontrow would act on first.

Metadata understanding for SharePoint library queries

Roadmap item 516044, launched with general availability listed as November 2025, introduces column metadata support for Copilot queries when a SharePoint library or folder is attached or its URL is provided. Copilot can use the library's own columns, such as status, client, region or document type, to return more relevant results. For a business that has invested in proper metadata on a contracts or projects library, this is the payoff. Prerequisite: a Copilot licence and a library with meaningful columns. Governance question: columns sometimes hold things the document body does not, such as an internal risk rating or a commercial note, so check that column content is as appropriate to surface as the files themselves.

Authoritative content for Copilot Search

Roadmap item 503553 is launched, general availability listed as September 2026. In the Microsoft 365 admin centre, under Copilot, then Search, then Authoritative content, an administrator can add and manage up to 100 SharePoint sites treated as authoritative, which lifts their relevance and ranking in Copilot Search. The 11 August release notes describe the SharePoint-side equivalent (roadmap 561323): marking a site as authoritative so its news, policies and updates are prioritised.

Prerequisite: admin centre access and a short, honest list of which sites hold the current policy, price list and procedures. Governance question: this is the best control in the set for an SMB because it addresses Copilot citing the wrong version, and it only works if those sites are kept current. Pair it with the stale content cleanup at /insights/sharepoint-stale-content-cleanup-before-copilot.

Connector hygiene checklist for an Australian SMB

1. 1Inventory every connector under Settings, Search and intelligence, Data sources, and name a business owner for each. A connector nobody owns gets removed.
2. 2For each connector, test the identity mapping rule with one user who should have access and one who should not. The 6 October change makes it editable in place.
3. 3Approve federated (MCP) connectors one source at a time, with the data owner's sign-off, because there is no index to review afterwards.
4. 4Run the permissions checklist at /insights/sharepoint-onedrive-permissions-checklist-before-copilot before scanned-PDF matching reaches the tenant, HR, finance and legal libraries first.
5. 5Label the libraries that hold scanned contracts, supplier forms and identity documents, using the recipes at /insights/copilot-dlp-sensitivity-labels-policy-recipes, so a label rather than luck decides what Copilot can quote.
6. 6Set authoritative content sites in the admin centre and diarise a quarterly check that they are still current.
7. 7For Power BI and Dataverse, list the models and tables certified for Copilot to answer from, and review row-level security and security roles on those specifically.
8. 8Decide, in writing, who may connect a CRM such as HubSpot to Researcher and for which user group.
9. 9Re-run the list whenever a connector is added or an administrator leaves.