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.