Research mode — investigate deeply like a McKinsey consultant

Introduction

Ask Claude a question, and you get an answer. But the truly valuable issues — "should I enter this market?" "why are sales declining?" — cannot be solved with a single answer. They require a multi-step investigation: asking the right questions, gathering multiple sources, analyzing, and then drawing conclusions.

That's when Research mode shines — and it will shine even brighter if you borrow the mindset of a McKinsey consultant. This article — the 16th in the series — combines Claude's multi-step investigative capabilities with structured analytical discipline, so you can investigate an issue like a mini consulting team.

What is Research mode

Research mode is Claude's mode for conducting a multi-step investigation: breaking down questions, finding and cross-referencing multiple sources, and then synthesizing them into a sourced report. Unlike a one-off Q&A — it’s like assigning a research assistant an entire project.

• Multi-step: Claude chains searches and analyses on its own instead of answering immediately.

• Multiple sources: cross-referencing information from various places instead of a single answer.

• Sourced: so you can verify — a must for anything serious.

One question is not enough for a big issue

The most common mistake is to ask a big question directly and expect a concise answer. "Should I open a coffee shop?" is a question that cannot be answered properly in one go — it encompasses dozens of smaller questions.

• The big question needs to be broken down into several smaller, answerable questions.

• Each small question requires its own data, and cannot be lumped together at once.

• Rushing to answer a big question often results in vague, useless advice.

Consultant mindset: ask the right questions before seeking answers

What makes a good consultant is not knowing a lot, but knowing how to ask the right questions. They start with a hypothesis and a core question, then let the entire investigation serve to validate that hypothesis.

• Articulate the core question sharply: precisely what decision you need to make.

• Formulate an initial hypothesis to guide the investigation, rather than gathering data aimlessly.

• Always ask "so what" — data is only valuable if it leads to a decision.

MECE: divide the problem without overlap or gaps

MECE is a classic McKinsey tool: dividing a problem into non-overlapping and exhaustive parts. It helps make your investigation both concise and comprehensive.

• No overlap: branches do not step on each other, preventing duplicated efforts.

• No gaps: when combined, branches must cover the entire issue.

• Ask Claude to propose a MECE breakdown, then review it for overlaps or gaps.

The 5-step investigation process with Research mode

A structured investigation goes through five steps, and Claude accompanies you at each step:

• Definition: finalize the core question and scope.

• Hypothesis: state a predicted answer to provide a direction for validation.

• Collection: have Claude gather data for each MECE branch, along with sources.

• Analysis: look for patterns, contradictions, and evidence supporting or refuting the hypothesis.

• Synthesis: draw conclusions and recommendations, not just a pile of data.

Force Claude to source and self-critique

This is the crucial part of the investigation: AI can confidently fabricate data and citations. A consultant never presents figures without sources — neither should you.

• "Each number must be accompanied by a source; wherever there is uncertainty, clearly state that it has not been verified."

• "Critique your own conclusions: what evidence could refute them?"

• Cross-check important data with its sources before making decisions.

From Data to 'So What': Summarizing Recommendations

The value of a consultant lies not in gathering data, but in the question "so what." A pile of data is not an answer; clear recommendations based on data are.

• After each finding, ask: what does this mean for my decision?

• Consolidate findings into 2-3 key messages, not ten disparate pages.

• Conclude with a conditional recommendation: what should be done, under what assumptions.

A Real Investigation Run: Example

Situation: you are considering opening a small café in a new residential area.

• Definition: "Is there enough demand in this area for a small café to be profitable within 12 months?"

• MECE: divide into demand (residents, habits), supply (local competitors), costs (rent, operations).

• Investigation: ask Claude to gather data for each branch with sources, marking any unverified information.

• Synthesis: "Should try on a small scale if the rent is below X, as demand is sufficient but competition is high."

5 Traps When Investigating with AI

• Trusting the data Claude provides without asking for sources — easy to fall for fabricated data.

• Asking a big question all at once, without breaking it down, leading to vague advice.

• Only seeking evidence that supports existing beliefs (confirmation bias).

• Stopping at a pile of data, forgetting the "so what" step to generate recommendations.

• Not stating hypotheses, allowing the investigation to drift aimlessly and become diluted.

Results You Will Get After This

• The ability to turn a big question into a structured investigation, rather than vague advice.

• More reliable conclusions because they are sourced and have undergone self-critique.

• Consultant mindset: ask the right questions, divide MECE, and always conclude with "so what."

First Steps for In-Depth Investigation This Week

• Choose a real decision you are uncertain about, and write it as a concise core question.

• Ask Claude to break that question into a MECE tree, then self-reflect to see if there are overlaps or gaps.

• Use Research mode to gather data for each branch, requiring each number to be accompanied by a source.

• Self-synthesize into 2-3 key messages and a conditional recommendation.

Conclusion

Research mode gives Claude the legs to conduct multi-step investigations; the consultant mindset gives it the head to investigate in a structured way. Ask the right questions, divide the problem MECE, enforce sourcing, and conclude with "so what" — that is how to turn a vague question into a recommendation you can bet on. The hasty questioner will receive advice anyone can give; the disciplined investigator will have their own answers, backed by evidence. Investigate, don’t just ask.