Research and Document Analysis Prompts

Research and document analysis represent one of the most high-value use cases for Claude. Whether you are synthesizing a literature review, extracting insights from a contract, or comparing policy documents, the quality of your prompts directly determines the quality of your analysis.

Large Document Analysis

When working with a single large document, your prompt needs to direct Claude's attention effectively. Rather than asking "summarize this document," give Claude a specific analytical lens.

Weak prompt:

Summarize this report.

Strong prompt:

Read the attached annual report. Identify the three most significant risks mentioned in the management discussion section. For each risk, extract: (1) how the company frames the risk, (2) any quantitative data provided, and (3) the proposed mitigation strategy. Present your findings in a structured format.

The strong version tells Claude what to look for, where to look, and how to present findings. This is the difference between a generic summary and a targeted extraction.

Key Point Extraction

For extracting key points, be explicit about the extraction criteria:

• By importance: "Extract the five most consequential claims the author makes, ranked by potential impact."
• By type: "Extract all recommendations, all statistics, and all named stakeholders."
• By relevance: "I am evaluating this vendor for a data security role. Extract all claims related to security, compliance, and data handling."

Multi-Source Synthesis

Synthesizing multiple documents is where XML tags become essential. When you send Claude several documents at once, XML tags prevent it from confusing sources and allow it to reference them precisely.

Using XML Tags for Multi-Document Research

Structure your prompt so each document is clearly delimited:

This template gives Claude clear source identities, a structured task, and explicit citation instructions. The <synthesis_task> block separates analytical instructions from raw data, which reduces the chance Claude will conflate them.

Agreement and Divergence Mapping

A powerful pattern for multi-source research is asking Claude to map positions explicitly before synthesizing:

Before writing your synthesis, create a table with the following columns: Topic | Document 1 Position | Document 2 Position | Document 3 Position | Assessment. Fill in the table first, then write the synthesis based on it.

This forces Claude to do the comparative work explicitly, producing a more rigorous output than asking for synthesis directly.

Citation and Source Referencing

Claude does not fabricate citations when you provide the source material directly. However, you must instruct it on how to cite:

Vague instruction: "Reference your sources."

Precise instruction: "When making a claim that comes from a specific document, cite it inline using the format (Doc ID, paragraph number if identifiable). If a claim synthesizes multiple sources, list all relevant document IDs."

You can also ask Claude to flag uncertainty:

If you are uncertain whether a claim is supported by the provided documents or are drawing on general knowledge, mark it with [inference] so I can verify it.

Research Methodology Prompts

For systematic review tasks, structure your prompt around a methodology:

Literature Analysis Pattern:

I am conducting a systematic review of [topic].

Criteria for inclusion: [your inclusion criteria]
Criteria for exclusion: [your exclusion criteria]

For each of the following abstracts, determine:
- Include or exclude, with brief justification
- If included: study type, sample size, primary outcome, key finding

[Abstracts follow]

PICO Framework for Medical Research:

Analyze this study using the PICO framework: Population (who was studied), Intervention (what was done), Comparison (what it was compared to), and Outcome (what was measured). Then assess the study's limitations as disclosed by the authors.

Summary

Effective research and analysis prompts share common traits: they use XML tags to organize multi-document inputs, they specify exactly what to extract and how to present it, they provide explicit citation instructions, and they ask Claude to be transparent about uncertainty. The investment in prompt structure pays off in analysis quality that would otherwise require significant human editing to achieve.