AI for Research, Writing, and Analysis

Where AI Adds the Most Value for Knowledge Workers

For most non-technical professionals, the highest-leverage AI applications are not exotic AI systems - they are better use of the conversational tools already available. Research compression, first-draft writing, document synthesis, and structured analysis are the tasks where AI tools can save hours and where the failure modes are manageable with the right habits.

This module covers the specific workflows that work and the risks to manage carefully.

Research: Compression, Not Replacement

AI tools excel at compressing research: reading a large amount of text and extracting what is relevant to a specific question. What takes a human two hours to read and synthesize can often be summarized in two minutes.

What this is good for:

• Getting oriented in an unfamiliar domain quickly
• Identifying which primary sources to prioritize reading
• Synthesizing a set of documents you have already collected
• Generating a list of questions to investigate further

What this is not good for:

• Discovering recent information (training data has a cutoff)
• Finding information that was not in the training data
• Reliable citations - models frequently confabulate convincing citations that do not exist

The research workflow that works:

1. Use AI to orient yourself and identify key concepts
2. Use primary sources (academic papers, official documents, authoritative databases) to verify specific claims
3. Use AI again to help synthesize what you have confirmed from primary sources

Think of AI as an accelerator for the research process, not a replacement for it. You still need to find and read the authoritative sources - AI just helps you prioritize and synthesize them faster.

Writing: Drafts and Editing, Not Ghost-Writing

AI tools are most useful for writing as a thinking aid and drafting accelerator, not as a replacement for your voice and judgment.

Where AI genuinely helps:

• Overcoming blank-page paralysis: a rough draft to react to is easier to improve than starting from scratch
• Reformatting: converting notes into prose, prose into bullets, paragraphs into executive summary
• Tone adjustment: making something more formal, more conversational, more concise
• Finding gaps: asking "what questions does this document not answer?" surfaces blind spots
• Alternative phrasings: "give me three different ways to say this sentence"

Where AI writing assistance fails:

• Original analysis: AI can generate text that sounds analytical but often produces conventional-sounding platitudes rather than genuine insight
• Your specific context: AI does not know your organization's situation, history, or the relationship dynamics that shape how a communication will land
• Accuracy of specific claims: AI-generated factual assertions in a document require the same verification discipline as any other AI output

The writing workflow that works:

1. Outline the main points yourself - this is the intellectual work
2. Use AI to generate a rough draft from the outline
3. Edit heavily - treat the AI draft as raw material, not finished work
4. Have another human review anything that will be seen externally

Document Analysis: Where AI Is Most Reliable

Analyzing documents you have provided is the AI task with the most reliable outputs. The model is working with information you gave it, not drawing on external knowledge it may have wrong. The main risk is summarization errors - places where the model distorts or omits something important.

High-value applications:

• Summarizing long documents or reports
• Comparing two documents and identifying differences
• Extracting structured information (dates, names, numbers) from unstructured text
• Answering specific questions about a document's content

Verification practice for document analysis:

• For summaries, spot-check 3-5 claims against the original
• For data extraction, verify a sample of extracted values
• For comparison tasks, read the specific sections the AI claims differ

Structured Analysis: Prompting for Frameworks

AI tools can apply structured analytical frameworks reliably when the framework is described clearly in the prompt.

Example prompts:

• "Apply a SWOT analysis to the following business situation: [description]"
• "Identify the three most significant risks in the following proposal, and for each, suggest one mitigation strategy: [document]"
• "Compare these two vendor proposals on the following dimensions: price, delivery timeline, support terms, and customization flexibility. Present as a table."

The output quality depends entirely on the quality of your question. Vague questions produce generic framework applications. Specific questions constrained to specific documents produce useful analysis.

The Risk: Outsourcing Your Thinking

The most significant risk in AI-assisted knowledge work is not factual errors - it is the gradual outsourcing of the thinking itself.

Using AI to draft means you skip the cognitive work of finding the right words to express an idea - and with it, the process of clarifying the idea. Using AI to analyze documents means you skip the close reading that surfaces things the AI will miss or mischaracterize.

The mitigation: use AI to accelerate your thinking, not to replace it. The thinking - the judgment, the synthesis, the decisions - must remain yours. AI produces material to react to, not conclusions to adopt.

Where to Go Next

The final module of Phase 2 gives you a systematic framework for evaluating AI tool outputs across any task type - taking the specific insights from these four modules and turning them into a repeatable practice.