Start With Research, Not Writing
When a task feels big, the first mistake is asking for the finished thing immediately: “Write my sales page,” “Plan my trip,” or “Compare these vendors.” The AI has to guess what matters, choose a structure, write the answer, and check itself all in one pass. That usually produces something polished but shallow.
Instead, begin with a research pass. Research simply means gathering useful facts, constraints, examples, and questions before deciding what to say. For example: “I need to compare three meal delivery services for a family of four. Ask me what you need to know, then summarize the main factors I should compare.” This gives the AI a narrower job: understand the situation. The tradeoff is that it takes one extra step, but it prevents the most common failure: a confident answer built on missing context.
Turn Research Into an Outline
Once the AI understands the material, do not let it jump straight to the final version. Ask for an outline first. An outline is the skeleton of the answer: sections, order, key points, and what each part should accomplish. This is where you can catch problems cheaply, before the AI spends time writing paragraphs in the wrong direction.
A useful prompt is: “Using the research above, create an outline for a one-page recommendation. Include the main sections, the point of each section, and what evidence belongs there.” For a work email, ask for the subject line, opening, decision needed, supporting details, and next step. For a project plan, ask for goals, milestones, risks, owners, and open questions. If the outline feels too broad, ask the AI to cut it. If it misses something important, add it now. The outline becomes the agreement the draft should follow.
Draft From the Outline Only
After the outline is right, ask for the draft. This step works better because the AI is no longer inventing the plan while writing the language. It is following a map you already approved. Your prompt can be simple: “Write the first draft using this outline. Keep it clear, practical, and under 600 words. Do not add new sections unless something is truly missing.”
That last sentence matters. AI often expands the scope when it tries to be helpful. For a beginner, the safest habit is to keep each pass disciplined. Research gathers material. Outline organizes it. Draft turns it into readable prose. If the draft is too formal, too long, or too vague, revise the instruction: “Rewrite for a busy manager,” “Make it sound like a helpful colleague,” or “Replace general claims with specific examples from the research.” Do not restart from scratch unless the structure is wrong.
Review Like a Separate Job
The final pass is review. Review means asking the AI to inspect the draft for problems instead of improving everything at once. This role is different from writing. A good reviewer looks for weak logic, missing facts, unclear wording, unsupported claims, and places where the reader may get confused.
Use a prompt like: “Review this draft as a skeptical but helpful reader. List the top five issues, explain why each matters, and suggest a specific fix. Do not rewrite yet.” Then choose which fixes you want and ask for a targeted revision. This avoids another common failure: endless rewriting that makes the piece smoother but not better. The whole workflow is simple: first understand the task, then shape it, then write it, then test it. Bigger requests improve when the AI has one job at a time.
Key takeaways
- Do not ask for a finished answer first when the task has many moving parts.
- Use a research pass to gather facts, constraints, examples, and missing questions.
- Approve or revise an outline before asking for a full draft.
- Tell the AI to draft from the outline and avoid adding new sections without reason.
- Use review as a separate pass focused on finding problems, not rewriting everything.
- For larger tasks, one extra step early usually saves several messy revisions later.