Three years into grad school, no one taught me how to write a paper—until I straightened out this workflow
Let me start with an unpleasant but very real situation: the vast majority of grad students are pushed into writing papers without ever receiving systematic writing training. Undergraduate essays and academic papers are fundamentally two different things, yet no one teaches us how to structure an abstract, how an introduction differs from an abstract, why formatting has to use LaTeX, or how exactly to respond to reviewer comments.
So the common scene becomes this: formatting in Word until you break down, struggling for ages and failing to produce an English sentence, redoing figures and tables over and over, and receiving reviewer comments without knowing what the first sentence should say. Most of the pain of writing a paper does not come from "the research wasn't done"—it comes from these rules that no one explains clearly.
This article organizes the experience I've accumulated from years of stumbling through pitfalls, from choosing a format all the way to replying to reviewers. If something can be copied directly, just copy it.
First, get clear: what exactly is a paper?
Ultimately, a paper is a stage-by-stage summary of a research project, presenting it systematically and rigorously in written form. The reasons for writing one vary widely—promotion, academic exchange, and for students, more often to meet graduation requirements. But strip away these external motives, and the core of a paper never changes: explain what you have produced clearly and rigorously, in a way others can reproduce.
Understanding this is important. It determines the criteria for judging everything you do later—any operation that makes "explaining it clearly" smoother is worth doing; any lazy writing that only saves you effort will eventually fail with reviewers.
Formatting: stop fighting with Word
The first thing to correct is tool choice. It is not recommended to write a paper from scratch in Word; once formatting needs adjustment, it becomes very painful, especially for long documents with equations, references, and cross-references. A more stable approach is to use the LaTeX source files provided by the journal or conference and make only small changes on that basis. The template already configures fonts, line spacing, heading levels, and figure/table floating rules, so you just fill in content, and formatting trouble can be cut by more than half.
There is also a lazy but efficient method: do not force yourself to write against a blank page. First look through the .tex source files of papers already accepted by the journal, see how others arrange the structure, write equations, and word figure captions, and then modify them against the official template. Finished work by predecessors is the best formatting textbook, much faster than figuring out template parameters yourself. Many research groups or courses also directly organize source-file packages from various publishers; if you can get them, don't waste the opportunity.
Abstract: ten sentences to carry the front
The abstract is the condensed version of the whole paper, and readers often decide whether to keep reading based on this paragraph alone. It generally contains three parts: a review of existing work, what this paper does, and experimental results.
Keeping it to around ten sentences is appropriate, and the rhythm can be arranged like this:
1. Point out the problem and explain why it matters;
2. Explain what predecessors did and where they fall short;
3. State what this paper does;
4. List the paper's main contributions;
5. Briefly describe the experimental setup and key results;
6. End with a sentence about significance, but don't resort to slogans.
Behind this order is a simple logic—first get readers to agree that this is worth doing, then tell them how far you have taken it.
Title and keywords: short, accurate, easy to search
The title's job is to be found and clicked, so it should be attractive, clear in meaning, and use searchable wording. Don't make it too long; forty to fifty letters is enough. Put the most core method and object into the main title.
Keywords supplement the abstract: three to five keywords, each one to three words. Piling on more dilutes the focus; more is not necessarily easier to search.
Introduction: expand each abstract sentence into a paragraph
The introduction is the bridge before the main text, responsible for leading readers from "this matters" to "how you plan to solve it." It can share the same rhythm as the abstract—one sentence in the abstract can be expanded into one paragraph in the introduction. Keep each paragraph to five to ten sentences, and don't cram everything into one paragraph.
When writing the introduction, there is a technique: the same point can be confirmed repeatedly from several angles, but the wording must change; don't repeat yourself. Expression should be concise and appropriate, and not a single spelling or grammar error should remain—such low-level mistakes directly lower reviewers' trust in the whole paper.
Literature review: praise generously, criticize politely
A review demonstrates respect for prior work. It cannot just list references; it should introduce them by category and evaluate their strengths and weaknesses. Two senses of proportion must be handled well: strengths can be praised generously, while weaknesses should be fair and polite. Don't criticize peers into oblivion—that is planting mines for your own review. Also, don't pile citations into one huge block; introducing them in different places and by theme is more natural.
Hard rules in the main text
This part is the most tedious, but it most easily exposes professionalism. Here are a few points repeatedly emphasized:
· Don't use colloquial contractions or casual words: contractions such as haven't and don't should not appear in formal papers; don't put and at the beginning of a sentence; use words like easy and simple cautiously—what looks simple to you may not seem so to reviewers.
· Symbols need rules: when many mathematical symbols are involved, provide a symbol table for easy reference; the same letter in different fonts means different things, so don't change it arbitrarily; do not apply fancy font treatments to symbols or equations.
· Write equations as sentences: mathematical expressions are part of sentences, so include commas and periods where they should be included. If unsure, look at published papers' .tex files and learn how others write them.
· Keep pseudocode focused on the key points: the core algorithm is one of the paper's showpieces. When writing pseudocode, keep the key steps clear and omit irrelevant procedures.
· Experiments must be credible: list datasets completely and state sources clearly; the broader the coverage, the more solid the results. Do not present the same set of results repeatedly in multiple places. Internal comparisons should show performance changes caused by parameter changes; external comparisons should prioritize classic yet advanced methods as baselines; let figures and tables speak, put the most important comparisons in prominent positions, and provide textual interpretation.
· Figures and tables require care: always use vector graphics for figures; don't hand-draw them in drawing software. Generate them with tools such as MATLAB and Visio, otherwise they blur when enlarged. Figure captions should be longer, usually one to several complete paragraphs, not just three words. Table margins should align with the main text, and datasets should be sorted according to some rule, not left out of order.
· Manage references with LaTeX: formatting must be correct, equals signs aligned, and author-name style and title capitalization must follow the style guide.
Conclusion: wrap up in five sentences, don't repeat the abstract
The conclusion does not need to be long; about five sentences is enough. One key point: don't copy abstract sentences verbatim. The conclusion should be more specific than the abstract, settling on what your method actually brings and where it can go next.
If your English sentences aren't good, a clumsy but effective method works well
Non-native speakers all feel uncertain writing English. Two practical approaches: first, find well-written papers in your field, break down and analyze their sentence structures, and imitate them; second, write a sentence yourself, translate it into Chinese, then translate it back into English, and repeat until the meaning no longer changes between the two expressions—that shows the sentence's logic is self-consistent. Most grammar problems can be covered by these two methods.
Writing a paper for the first time: do it in three passes
Don't expect to get it right in one draft. In the first-draft stage, build from details toward the whole; if you can borrow a mature format, set up the skeleton first. In the revision stage, reverse the direction—go from the whole toward details, smooth out the logic before polishing sentences, and focus on improving readability. Only at the end should you check and polish, specifically hunting for problems and refining language. Separating "writing" and "revising" into different passes is much more efficient than agonizing while writing.
When you receive reviewer comments, get your mindset right first
Attitude is the first hurdle in replying to reviewer comments. Don't be confrontational or write long defenses; keep replies direct and not too long. Mark changes in the main text in blue so reviewers can see at a glance what you changed. Many rejections are not because the comments are difficult, but because the reply reads like an argument.
In this workflow, what really drains you is often repetitive work
The truly difficult part of putting the above techniques into practice is "the tedium of execution": searching for literature takes time, formatting adjustments are painful, English expression gets stuck, and aligning references is cumbersome. None of these tasks is difficult when separated, but piled together they create enormous consumption.
This is also why I recommend MuchSci Research Cloud Brain
—it is not a single-point tool, but integrates the capabilities needed across the entire research workflow into one platform. Its core advantages are two:
180 million paper database + natural language search:MuchSci's paper search assistant supports asking questions directly in natural language, such as "Find me five highly cited papers on protein structure prediction, sorted by citation count." It will automatically call search tools to run the search, sort results, and return complete metadata (abstract, authors, DOI, citation count, etc.). For literature it cannot find in the local database, it can automatically go online to fill in the information, saving you the trouble of manually searching across databases.
One-stop solution for the entire writing workflow:From literature search, academic translation, and writing polishing to formatting and typesetting, all steps are completed within one platform. When you are unsure about an English paragraph, use its academic translation capability to accurately turn the Chinese meaning into standard English, with proper nouns following field-standard translations; for repetitive work such as reference formatting and layout of the experimental section, hand it over to organize a first draft according to the target journal's format, and just verify the data yourself; from close reading of literature to final manuscript layout, it can basically be connected within one platform, eliminating the effort of exporting data back and forth among translation, writing, and formatting tools.
What it takes off my plate is the "standardization" and "transfer" layer; judging whether the research is explained clearly and whether the conclusions hold up still has to be done by me. But for grad students tormented by formatting and language, being able to hand off this repetitive work can already save a considerable amount of time.
Final thoughts
There is no mystique to paper writing; it relies on a set of rules previous generations have long summarized: leave formatting to templates, build the abstract with a rhythm, follow standards for figures and tables, manage references with tools, and revise in three passes. Once you straighten out this workflow and let tools take over the mechanical work that doesn't require your personal judgment, writing a paper turns from an anxious task into one that can move forward step by step.