AI for research and first drafts

AI as a research accelerator, not an oracle

The most common mistake when using AI for research is treating it like a search engine that always finds the truth. It doesn't. It generates plausible text based on patterns in its training data. That text can be wrong, outdated, or — most dangerously — confidently stated but fabricated.

Used correctly, AI is an excellent starting point: it gives you the landscape quickly, helps you know what questions to ask, and surfaces concepts and angles you might have missed. The verification work is still yours. Think of AI research output as "a knowledgeable colleague's first take" — useful but fallible.

Effective research prompts

These patterns consistently produce useful research output:

• "Give me an overview of X, then list the five most important things someone new to this topic should understand." Produces a structured orientation, not a wall of text.
• "What are the main competing views on X? What does each side argue?" Useful for complex topics where the right answer isn't settled.
• "What questions should I be asking about X that I probably haven't thought of yet?" One of the highest-leverage research prompts — it maps your blind spots.
• "Summarise this document and pull out: the main argument, the key evidence, and anything I should be sceptical about." Paste in a document for rapid critical reading.

The first-draft mindset

The value of a first draft isn't quality — it's overcoming the blank page. A bad draft you can edit is always better than a perfect draft that doesn't exist. AI produces first drafts faster than any human, which means your job shifts: from writing to editing and directing.

This is a genuine skill change. Many people find editing AI output uncomfortable at first because it doesn't feel like "their" work. Push through that. The finished piece is yours — the AI just handed you raw material to shape.

How to get better drafts

The quality of a first draft is almost entirely determined by the quality of the brief. Before you ask for a draft, give the model:

• The purpose of the piece (inform, persuade, instruct?)
• The audience (who reads it, what do they care about?)
• The desired tone (formal, conversational, direct?)
• The structure you want (sections or headings to include)
• Any key points that must appear

If you have notes or bullet points, paste them in and ask the model to turn them into a draft. Your notes anchor the content so the model can't drift into generic territory.

Iterating on a draft

Don't settle for the first output. The best workflow is draft → critique → revise. After the first draft, ask: "What is the weakest part of this draft? How would you improve it?" Then ask it to fix the weak parts. Two rounds of iteration usually produces something much closer to what you need than a single attempt.