Paper Writing & Structure.

The AI Writing Principle

AI should edit your writing, not replace it. The difference matters for quality, originality, and ethics. When AI writes from scratch, it produces competent but generic prose. It lacks your specific theoretical perspective, your interpretation of the data, and your voice. The result reads like a textbook summary -- technically correct, intellectually empty. When AI edits your draft, it tightens your arguments, catches logical gaps, improves transitions, and polishes grammar while preserving your ideas. The result reads like your best work, produced faster. The workflow: You write a rough draft (messy, incomplete, full of notes-to-self). AI restructures, clarifies, and polishes. You review, revise, and add the intellectual content AI cannot supply. This cycle repeats until submission quality is reached.

Section-by-Section Strategy

Each section of a research paper has different AI affordances: **Abstract (write last, edit heavily)** ``` Prompt: "Edit this abstract to be exactly 250 words. Ensure it covers: background (1-2 sentences), gap/objective (1 sentence), methods (2-3 sentences), key results with numbers (2-3 sentences), conclusion/implication (1-2 sentences). Remove all filler." ``` **Introduction (AI outlines, you write, AI edits)** ``` Prompt: "I'm writing an introduction for a paper on {topic}. My argument moves through these points: 1. {broad context} 2. {narrowing to specific problem} 3. {what's been done} 4. {what's missing -- the gap} 5. {how our work addresses the gap} Review my draft and identify: (a) paragraphs that don't serve the narrative, (b) missing transitions, (c) claims that need citations, (d) where the argument loses focus." ``` **Methods (AI generates boilerplate, you verify accuracy)** Methods sections are highly formulaic. AI excels here. ``` Prompt: "Write a Methods section describing: {procedure_details}. Use past tense, passive voice where appropriate. Include all details needed for replication: sample sizes, specific parameters, software versions, statistical tests with justifications." ``` **Results (you write, AI formats)** Results should come from your analysis code output, not from AI generation. Use AI to format the numbers into readable prose. ``` Prompt: "Convert these statistical results into a Results section: {paste_statistical_output} Follow APA format for reporting statistics. Include effect sizes. Do not interpret -- just report." ``` **Discussion (most human-dependent section)** The discussion is where you make your intellectual contribution. AI can help structure it, but the interpretation must be yours. ``` Prompt: "Review my Discussion section. Check that I have: 1. Restated key findings (without repeating Results) 2. Interpreted findings in context of existing literature 3. Addressed alternative explanations 4. Acknowledged limitations honestly 5. Stated implications (practical and theoretical) 6. Suggested future directions that are specific and actionable" ```