MPUG Editorial · September 8, 2026 · ~4 min read · Category: AI & Copilot

Copilot Researcher model selection arrived in the Microsoft 365 Copilot release notes dated 25 August 2026, and it is the kind of change that is easy to scroll past and then quietly benefit from. When you add the Researcher agent to a Copilot Chat prompt, you can now choose which supported Researcher model and mode runs. For project managers, that matters because research is where a surprising amount of project time disappears — vendor comparisons, options analysis, regulatory reading before a gate review — and until now the model doing that work was chosen for you.

What changed

According to Microsoft’s release notes, Copilot Chat now includes an option to pick from available Researcher models and modes when you add the Researcher agent to your prompt. Previously, users had no direct control over which Researcher model was used during chat interactions, which Microsoft describes as limiting customisation.

The mechanics are straightforward:

1. Start a new chat and add the Researcher agent to your prompt.
2. Select the model you want from the model selector menu.
3. Continue the conversation using the chosen model.

Microsoft lists the feature for Windows and web. Its stated rationale is that letting users select the model provides more control over the research process and improves result relevance by matching the model to the task. Microsoft’s own summary of the business impact is that teams can customise research workflows to better fit specific project needs.

Why it matters for project managers

The honest answer is that most project research is not one job – it is two, and they have different tolerances.

The first is breadth: a first-pass scan of a market, a list of candidate vendors, a quick read on whether a regulation applies. Here, speed wins. You are trying to decide whether a topic deserves an hour, and a fast model that gets you 80% of the shape is the right tool.

The second is defensibility: an options paper that a steering committee will approve or reject, a make-versus-buy analysis that sets a budget, evidence attached to a risk you are asking a sponsor to accept. Here, a heavier reasoning model earns its extra latency, because the cost of a plausible-but-wrong synthesis is a decision you have to unwind later.

Before this change, you got one setting for both. The practical discipline now is to decide which job you are doing before you prompt, rather than discovering after reading a thin answer that you needed the other mode. That is a habit, not a feature – but the feature is what makes the habit possible.

It also changes how you should treat AI research in your governance. If a business case cites AI-assisted analysis, “which model, in which mode” is now a legitimate question at a gate review, in the same way that “which data source” always has been. Teams building AI into delivery may find MPUG’s on-demand master classes on AI-assisted delivery governance useful here, and the PMI-ACP training path covers the iterative decision-making that pairs naturally with fast-then-deep research cycles.

What this does not do

Two caveats worth stating clearly, because AI release notes invite over-reading.

Model selection does not make Researcher output citable evidence on its own. It changes which engine produces a synthesis; it does not verify the synthesis. Whatever your practice was for checking AI research against primary sources, keep it.

Nor is model choice a substitute for a good prompt. A precise question sent to a fast model usually beats a vague one sent to a slow model. If your research output is disappointing, the prompt is the more likely culprit than the model.

What to do next

• Open Copilot Chat, add the Researcher agent, and confirm the model selector appears in your tenant – availability is listed for Windows and web, and rollouts vary.
• Pick a default pairing for your team: a fast model for scans, a heavier model for anything going to a decision forum. Write it down; two lines in your team charter is enough.
• Add a line to your business case template recording which model and mode produced any AI-assisted analysis, alongside the sources.
• Keep your verification step. Spot-check Researcher output against primary sources – vendor documentation, the regulation itself, the actual release notes.
• If you are formalising how AI shows up in delivery, MPUG’s live events and knowledge base cover the governance patterns other PMOs are landing on.
• Log the PDUs. Time spent building AI research practice is legitimate professional development – track it in the PDU tracker rather than reconstructing it at renewal.

Frequently asked questions

Where do I find the Researcher model selector in Copilot Chat? Add the Researcher agent to your prompt in a new chat, then use the model selector menu to pick from the available Researcher models and modes, per Microsoft’s release notes.

Which platforms support Copilot Researcher model selection? Microsoft lists the feature for Windows and web.

Does choosing a different model change how Copilot accesses my data? Microsoft’s release note describes the change as model and mode selection within the Researcher agent; it does not describe a change to data access. Check your tenant’s Copilot configuration and Microsoft’s current documentation before assuming anything about grounding or permissions.

The bottom line

This is a modest control with a disproportionate effect on a task project managers do constantly. Match the model to the job, record which one you used when the output feeds a decision, and keep verifying against primary sources. That is the whole practice – and it takes about five minutes to adopt.

Source: Release Notes for Microsoft 365 Copilot – Microsoft Learn, 25 August 2026 entry. Copilot feature availability depends on licensing and tenant configuration and rolls out on Microsoft’s own schedule; verify the specifics for your organisation against Microsoft’s current documentation.