Microsoft 365 Copilot just got a new thinking partner — and most users are briefing it the wrong way.

You already use Copilot Researcher for deep research tasks. But since Microsoft added Anthropic’s Claude models — including Claude Opus — to the Researcher experience, the rules of engagement have changed. Not because the model is smarter. Because the architecture is fundamentally different now.

Most people still use Researcher like a fast search box. Type a question, get an answer. That worked when Researcher was a single-model assistant. It no longer fits what the product has become.

Here is what changed — and how you need to adapt.

---

What Microsoft Actually Changed in Researcher

Researcher is not just a search agent anymore. Microsoft introduced multi-model capabilities that run in parallel:

• Critique mode: One model generates a response, another reviews it for gaps and inconsistencies.
• Council mode: Claude (Anthropic) and GPT (OpenAI) produce separate reports, which are then compared and synthesized.

This means Researcher can now surface disagreements between models, flag where conclusions differ, and highlight where they converge. That changes everything about how you should prompt it.

---

How to Brief Researcher Like an Analyst — Not a Chatbot

Step 1: Define the decision, not just the topic

Bad prompt: “Research document automation vendors.”

Better prompt: “We are evaluating three vendors for document automation. First outline evaluation criteria, required internal sources, external validation points, risks, and open questions. Then generate the report.”

The first prompt gets you a summary. The second gets you a structured research process.

Step 2: Ask for explicit uncertainty

Polished language is not the same as confidence. Instruct Researcher to separate:

• Verified findings
• Inferred conclusions
• Unresolved gaps

This works especially well for vendor comparisons, security assessments, and policy analysis. You get synthesis with traceability — not just synthesis.

Step 3: Use Council mode on purpose

When a topic is ambiguous, politically sensitive, or fast-moving, Council mode is not just a feature demo — it is a validation layer. If Claude and GPT converge, your confidence increases. If they diverge, that is a signal to inspect the assumptions behind both o>>>utputs before deciding anything.

Step 4: Prompt for decision-support formats

Ask Researcher for outputs that are immediately actionable:

• A recommendation memo with three options and trade-offs
• A risk register with likelihood and impact columns
• An executive brief plus a detailed appendix
• A “what would change this recommendation” section

These formats are more useful than a generic long report because they turn research into a decision, not a reading assignment.

---

The Mindset Shift That Matters

Better models help. But better orchestration starts with better instructions.

The new reality is not: “Claude makes Researcher smarter, so I can do less.”

It is: “Researcher is becoming a multi-model reasoning system. I need to brief it like an analyst and review it like an editor.”

The human role remains irreplaceable: framing the business context, defining what a good answer looks like, and making the final call. What changes is how deliberately you hand over the task.

---

Start with your next Researcher prompt. Add a goal, an audience, and a decision. Then ask for a plan before the report. That one change will show you what the new Researcher is actually capable of.

---

More resources:

Microsoft 365 Blog – Introducing Researcher and Analyst

Anthropic – Claude in Microsoft 365 Copilot ·

Microsoft Community Hub – Multi-model intelligence in Researcher

Talk to us at HanseVision about your AI needs and requirements (OnPremises, Hybrid, or Cloud with M365 Copilot / Studio). Find my Calendar here

Also keep in mind that Copilot Researcher with Anthropic Models currently (April 2026) breaks EU Data Boundaries and could conflict with your compliance rules and regulations. Check out my Blogpost.