Prompting Claude for Analysis and Research Tasks

1.Why the task type changes the prompt

Analysis, research, drafting and brainstorming all start from the same habits. Explain the ask clearly. Include the context Claude has no way of guessing. Break large requests into steps. Anthropic's support guidance puts the core rule plainly: make sure the prompt contains as much context as possible, and write it as if the reader had no background on the topic.

The task type decides where most of your effort goes. The AI Fluency framework gives you a way to see this. Its Description competency has three parts. Product Description defines what the output should be. Process Description guides how Claude approaches the work. Performance Description shapes how Claude behaves while you collaborate.

View each task type through that lens. Analysis and research bring information in, so process does most of the work: which material Claude works from, which dimensions it examines, and how it stays tied to evidence. Drafting sends information out to a specific audience, so the product carries the weight: voice, length, required elements. Brainstorming needs breadth before judgement, so the prompt has to ask for range and hold evaluation back for later. This page covers the two information-in task types.

Sources12

2.Analysis: ground it, focus it, let it say "not enough"

An analysis can only be as good as the material behind it, so give Claude the actual report rather than a description of it. With long inputs, position matters. Anthropic's prompting guide says to put long documents near the top of the prompt, above the question, the instructions and any examples. For long-document work it also recommends asking Claude to quote the relevant passages before it starts the task, which keeps it on the content that matters.

Next, name the dimensions you want examined. "Analyze this report" leaves Claude to guess what you care about. The M&A example in Anthropic's guide to reducing hallucinations names three focus areas. It also gives Claude a set phrase to use when the report doesn't support a conclusion:

As our M&A advisor, analyze this report on the potential acquisition of AcmeCo by ExampleCorp.

<report>
{{REPORT}}
</report>

Focus on financial projections, integration risks, and regulatory hurdles. If you're unsure about any aspect or if the report lacks necessary information, say "I don't have enough information to confidently assess this."

Many people reach for a persona first. Anthropic's prompt engineering article suggests something more direct: "Analyze this investment portfolio, focusing on risk tolerance and long-term growth potential" instead of assigning a role. The same thinking applies when you want a balanced view. Ask for the angles you need, such as supporting evidence and counterpoints, or several perspectives. Extended thinking exists for this kind of reasoning. Anthropic describes it as letting Claude explore multiple angles before it reaches a conclusion.

Finally, permission to say "I don't know" is part of the analysis prompt itself. Anthropic calls it a simple technique that can drastically reduce false information. For compliance-style reviews, the guide goes one step further and splits the analysis into ordered steps that keep Claude tied to the text.

Sources3456

3.Research: context, scope, and the right capability

Research prompts fail in a different way from analysis prompts. The usual problem is a broad question with nothing about who is asking. In Anthropic's AI Fluency course for nonprofits, an executive director researching an expansion from Portland to Seattle gets useful results by explaining who they are, who they serve, and exactly what they need to know. The course's research exercise lists what such a prompt contains: the specific policy or legislation, the organisation's context, what you need to learn (such as funding implications or compliance requirements), and the time frame or geographic scope.

The second decision is which capability fits the question. Anthropic's support guidance separates three of them:

If the research should draw only on documents you supply, say so. Anthropic's guidance recommends explicitly telling Claude to use only the provided documents and not its general knowledge, and asking it to cite quotes and sources for each claim so the answer can be audited.

Plan for verification from the start, because it is part of how a research task is done. The nonprofit course says to flag specific claims for verification, notice gaps and question how recent the information is. A good research prompt makes this easier by asking for sources up front.

Sources746

Exam traps

Each one states something that sounds right. Open it to see what is actually true.

1. 1.Giving Claude an elaborate expert persona is the strongest way to improve an analysis prompt.Why is that wrong?

   Naming the perspective and focus areas you want often works better than a role, and overly specific roles can make Claude less helpful.

   Covered in Analysis: ground it, focus it, let it say "not enough"

2. 2.A research prompt only needs the question, and adding details about yourself just biases the answer.Why is that wrong?

   Context about who you are, who you serve and what you need is what makes research output relevant and actionable. Generic questions get generic answers.

   Covered in Research: context, scope, and the right capability