daBongo LMS AI Training Courses

← Back to Course

Everyday Claude – Core Habits for Productive AI Work

Lesson 4: Claude for Writing, Research, and Analysis Tasks

Lesson Objectives

By the end of this lesson, students should be able to:

  • Apply a three-step workflow for writing and editing tasks with Claude
  • Use Claude effectively for research synthesis without overrelying on unverified claims
  • Frame analysis and review requests to produce actionable output
  • Distinguish between tasks where Claude generates from scratch versus improves existing work

Lesson Content

Writing and editing – starting position matters.

There are two writing workflows with Claude:

Generate from scratch: Give Claude the context, audience, purpose, format, and tone. Claude produces a first draft. Best for: boilerplate, routine communications, template-based content, first drafts you will heavily edit.

Improve existing work: Give Claude your draft and tell it what is wrong or what to improve. Best for: work with a specific voice, sensitive communications, content where accuracy matters, any situation where Claude's first-draft defaults would miss the mark.

For most professional writing, the improve-existing-work workflow produces better results than generate-from-scratch – because your draft contains your intent and voice, and Claude's job is to make it better rather than guess at both.

Research and synthesis – the verification discipline.

Claude synthesizes information effectively but cites from training data, not live sources. For research tasks, the effective workflow is:

  1. Ask Claude to synthesize what it knows about a topic and structure the key points
  2. Use that synthesis as a research map – identifying the areas and questions that need real sources
  3. Verify key claims independently before using them
  4. Ask Claude to help structure and write up your verified research

This "map then verify" workflow uses Claude's synthesis strength while protecting you from its accuracy limitation.

Analysis and review – asking the right question.

Claude performs well on analysis tasks when the question is specific and the evaluation criteria are clear. Compare:

  • "Analyze this proposal." – Produces generic observations.
  • "Review this proposal for a technical audience and identify: (1) the three strongest arguments, (2) the two weakest or unsupported claims, and (3) one question a skeptical reader would likely raise that the proposal does not answer." – Actionable, specific, reviewable.

For document review, give Claude the document (pasted text or uploaded file) and specific evaluation criteria. For decision analysis, provide the options, the constraints, and the criteria that matter – then ask Claude to evaluate against those criteria.

Voice and accuracy checkpoints.

Two quality checkpoints apply to every AI-assisted work product:

Voice check: Does this sound like you (or your organization)? Claude defaults to its own stylistic range. For anything going out under your name, read it aloud. If it does not sound like you, it needs editing.

Accuracy check: Is every factual claim you will rely on verifiable? Especially for numbers, dates, names, technical specifications, and any claim with real-world consequences.

Practical Example

A policy analyst uses Claude for a briefing paper.

Workflow: he first gives Claude a structured prompt asking it to synthesize the current debate on a policy topic.

He uses the synthesis to identify five key questions needing real sources.

He researches those sources manually, then gives Claude his notes and asks it to draft a structured briefing with a specific word limit and section format.

He then edits for accuracy and voice.

Total time: two hours instead of five.

Claude did the first-draft synthesis and structure; he did the research and quality control.

Lesser-Known Tip

For editing tasks, you can ask Claude to explain its edits alongside making them: "Edit this paragraph for clarity and brevity, and explain each change in brackets after the edited version." This turns a routine editing task into a learning opportunity – over time, you will internalize the most common improvements and write cleaner first drafts.

Safety Notes

For any written output distributed under your name – reports, client communications, public-facing content – you are responsible for its accuracy regardless of how it was produced. AI-assisted authorship does not reduce your accountability for the content's factual claims. Apply the accuracy check before every distribution.

Practice Task

Take a document you have recently written – an email, a report section, or a brief – and run it through two workflows. First: paste it into Claude with "Improve this for clarity and brevity, keeping my voice." Second: paste a new version into Claude with "Review this for unclear arguments and any unsupported claims." Compare what each workflow surfaced. Note which type of feedback was more useful for your work.

Completion Check

You should be able to choose between generate-from-scratch and improve-existing workflows for any writing task, describe the "map then verify" research workflow, write an analysis prompt with specific evaluation criteria, and apply the voice check and accuracy check before distributing AI-assisted work.

Log in and enroll to access lesson quizzes.

← Back to Course

Scroll to Top