Undergraduate and graduate students starting out in research often feel lost when writing their first paper. It is hard to know where to begin or whether you are heading in the right direction. I felt the same way. Even with guidance, what works depends on the research and the situation, so some trial and error is inevitable. Still, I hope the approach below can make the process a little easier.

Before you begin writing in earnest, it helps to organize the experiments and claims scattered across your work. The attached slides provide a template for doing this.

Download the research planning slides (English PPT)

  1. Connect each contribution to its evidence. Start by writing down the contributions of your research. Then identify the table or figure that supports each one. Do more than paste in experimental results: explain which comparison in a table or which trend in a figure supports your claim. If possible, prepare the tables and figures as they would appear in the paper, with captions.

  2. Plan additional experiments and check the logic of your research. If you need an ablation study, break your method down into its components and decide what you can learn by removing or adding each one. If you need an analysis, first explain why it is needed. Be specific about the insight it will offer readers and how you will present the results in tables or plots.

    Creating this document is itself a way to organize your thinking. It lets you see at a glance what you are claiming, what evidence you have, and what you still do not know. You may discover that you have run many experiments without obtaining the evidence your central claim needs, or that you have results whose meaning you have not fully explained. This helps you find gaps in the research and its logic before you start writing. Once you have done this, use whatever writing process suits you. For a first paper, I recommend continuing with the steps below.

  3. Build the outline yourself. Choose the section and subsection titles based on your own understanding of the work. Under each heading, sketch what you want to say and insert the tables and figures from the slides. At this stage, focus on designing the flow of the paper rather than perfecting individual sentences.

  4. Use AI to write a full first draft. Working in the repository that contains your experimental results and code, use the outline to have AI produce a LaTeX draft. You do not need to meet the page limit immediately. It is often easier to include everything you want to explain and then shorten it. Writing more does not mean filling in results you do not have: keep verified findings separate from ideas you have not yet tested.

  5. Read the draft yourself and clarify the claims. Go through it sentence by sentence, remove unnecessary material, and sharpen the argument. Check whether the draft says what you intended and whether the experimental results actually support its claims. Even a convincing sentence needs revision if it goes beyond what your research has shown.

  6. Compare AI’s edits as you refine the wording and length. The next task to give AI is to soften overly strong language and bring the paper within the page limit. Do not simply accept the revised version: compare what it changed and what it omitted. “The point still comes across without this part.” “I already said this earlier.” “This is as far as the evidence allows me to go.” Reviewing these changes is an opportunity to learn how to cut text and calibrate claims. AI’s edits are not always improvements, so also check that necessary explanations have survived.

  7. Add citations and check their context. Ask AI to identify statements that need citations, search for relevant papers, and add the references. Look beyond the bibliography. Ask why this sentence needs a citation and why this particular paper was chosen. Open the cited paper to verify that it exists and that its content supports your statement. This is another opportunity to learn how your work connects to existing research.

  8. Read every sentence again for clarity. Once the paper is nearly finished, read it yourself from the beginning. If you find a passage difficult to read, other readers will probably struggle too. Preserve the meaning, but rewrite it in a way you can understand immediately. In my experience, AI-generated prose still does not always align with what people find easy to read. A sentence can be grammatical and sound academic yet take too much effort to understand. Look for a simpler way to express it.

  9. Review the PDF’s visual presentation. Check whether a paragraph ends with a single word on an otherwise empty line, whether text inside figures and tables is too small or too large, and whether unnecessary whitespace makes the page look sparse. Work through these details to improve the reading experience as well as the content.