How to Write Gooder

Nov 25, 2025 16 min read

After publishing “How to professor”, several people said they found it helpful, and asked whether I had a similar post on writing. Luckily, we have held an annual writing workshop in the lab for the last few years, so there already was a presentation. This article is the text version of that presentation. As with the previous post, everything here is based on my own experience as an academic writer, with some grounding in prior work, so take it with a grain of salt. However, if it can help you improve your writing and avoid some of my mistakes, go ahead.

Before we start, though, let’s address the elephant in the room. Some people have asked whether we still need to know how to write in the age of LLMs. My answer is an unambiguous “yes.” You should, of course, use all the tools at your disposal, e.g., ChatGPT to turn your notes into a continuous text, Quillbot to make your text more fluent, or Grammarly to improve the grammaticality. However, all of those tools are ML applications, and the old adage of “crap in, crap out” holds here as well. Starting out with a well-written piece will yield better results than leaving it to the tools to fix some sloppy copy. Besides, some LLM tools have their own flawed writing tendencies, such as overusing superlatives and empty phrases (“let’s delve into this fascinating topic”). They also tend not to be as good at structural aspects as human editors. Knowing how to write well will help you spot inconsistencies in LLM suggestions. More importantly, though, they do not reflect your voice, and that can leave the writing feel generic and anodyne. They do, however, make decent critics for early drafts to help you spot what is missing.

Scientific Writing

The Importance of Good Writing

“Good writing is like a window pane.”
—George Orwell

Scientific writing is a fundamental skill for researchers and academics. Whether you are authoring a research paper or drafting a grant proposal, writing is crucial for clear and effective communication. No scientific paper is just the results: A surprising amount of academic writing is also storytelling. You need to frame the results and make the reader care about them (anecdotally, I have seen rejected papers get accepted after improving just the writing).

Good writing differentiates successful communication from misunderstandings. Even small things, like misplaced words or punctuation errors, can dramatically alter meaning. For example, consider the headline: “Rachael Ray finds inspiration in cooking her family and her dog.” Poor dog… Adding prepositions like “with” and “for,” or simply using commas, can prevent such unfortunate interpretations. Clear writing helps avoid confusion.

Many review forms specifically ask how readable a submission is. That might seem like a formality, but getting a low score here will likely obfuscate any strengths of your work in other areas.

Know Your Audience

Scientific writing should be clear to any well-prepared reader. First impressions matter: You win or lose the reader on the first page, most likely already in the abstract. Within the first few lines, you either capture their attention and make them want to read on – or you lose their interest (and potentially the paper acceptance). Engrossed reviewers will be much more inclined to rate your paper highly. They may even forgive minor details that are off later on. If you start out on the back foot instead, reviewers may feel you are wasting their time and will be much more critical of any minor detail.

With that in mind, we should tailor our writing to fit our audience: a scientific paper and a blog post have different audiences, and your writing should take that into account. Independent of the topic, any audience falls somewhere on the “What-So What” curve, i.e., how much they care about the technical details (What) vs. the reasons and implications of the work (So What). Your co-authors will care much more about the What than the So What (they know why you did it). For most other audiences, the importance of the So What increases. The depth and complexity of information you’ll include differ greatly from what you’d present to a class of high-schoolers, a peer group, or grant reviewers (which might not be professional peers). Understanding your audience helps you decide how much background information to provide, how much jargon is acceptable, and how much you need to invest to make them care about the problem. People in your field are intrinsically motivated (otherwise, they would not be in that field). Grant reviewers might not. Depending on the type of grant, these might not be people in your field, or even academics at all.

For any audience, though, make sure to reduce their cognitive load: Suspense is great in fiction, but in academic writing, spoilers are completely okay. Do not leave readers guessing what the paper is about; tell them upfront. Keeping track of 5 different scenarios that a research paper could evolve into is taxing even for experts. Save them the trouble and let them focus on the content.

I have often heard the advice not to make your audience feel dumb. Most people, myself included, understand that to mean not to oversimplify matters. However, that is probably not correct. I have never seen an audience dislike a paper just becasue it explained a complex matter simply: Nobody feels dumb if they understand what’s going on in a paper. What does make people feel dumb is if they don’t get what the paper is about. In a few cases, that might be due to their limited knowledge. In the overwhelming majority of cases, though, it’s due to bad writing.

“‘All you have to do is write one true sentence. Write the truest sentence that you know.’ So finally I would write one true sentence, and then go on from there. It was easy then because there was always one true sentence that I knew or had seen or had heard someone say.”
—Ernest Hemmingway

We often think of writing as an artistic expression that requires some sort of innate talent. And while great fiction writers certainly bring their own style to the table, even they will readily admit that good writing is mainly one thing: practice. (Case in point: despite his adventurous lifestyle, Hemingway slavishly followed a daily writing routine and rewrote often.) The good news is that means nobody is automagically good at it. It also means anyone can learn it. Even if English is not your native language, you can craft good scientific texts by following some simple rules of thumb and practicing them until they become second nature. They might not win a Pulitzer Prize, but they will get your message across and keep your readers engaged. All it takes is some simple rules, repeated improvements, and lots of practice.

I have been writing almost daily for about 25 years. I have published over 150 articles, two textbooks, and have a forthcoming trade book. I like writing. And yet, the first draft I write is still gonna be dogwater. I have accepted that and now try to get it out of the way as soon as possible. It’s better to have a mediocre first draft you can improve than wait forever for the perfect inspiration. And with a few basic guidelines in mind, you can make that first draft at least a good starting point.

Rules of Thumb

Short Sentences are more effective and easier to understand. If a sentence extends beyond two lines, split it into smaller ones.
Subject Placement: Introduce the subject early in the sentence to clarify who’s doing what, and put it close to the verb.
Full Verbs: Use full verbs rather than nominal constructions.
Active Voice helps with all of the above. Instead of saying, “Measurements were taken,” use “We measured.” Many of us were taught to use the passive voice in formal writing because it is considered more “objective.” In practice, it hides who did the work and also offloads responsibility. Be bold: you put all this effort into your work, stand for it!
Information Structure: Introduce new concepts at the end of the sentence and elaborate on them immediately after. (It’s okay to break the active voice rule above to help with this aspect.)
Clear Referents: avoid dangling pronouns/starting sentences with a pronoun: “This enables us” ⟶ “This method enables us.”

Avoiding Fluff

Eliminate unnecessary words or phrases to make your text clearer. Here is an incomplete list for easy find-and-replace:

at that point in time ⟶ then
at this point in time ⟶ now
in the event that ⟶ if
until such time as ⟶ until
on account of ⟶ because
in the majority of cases ⟶ mostly
has the capability of ⟶ can
in spite of the fact that ⟶ despite
in the final analysis ⟶ finally, ultimately
a large percentage of ⟶ most
owing to the fact that ⟶ because
need to be established ⟶ requires
give consideration to ⟶ consider
with the exception of ⟶ except
it would thus appear that ⟶ apparently
in order to ⟶ to
we aim to do X ⟶ we do X

Illustrative Examples

Consider this before-and-after example:

Before: “By investigating the characteristics of bidirectional embeddings, it is desired that new future roles for this approach will be determined.”
After: “Our investigation of bidirectional embeddings will reveal new roles for this approach.”

The revised sentence is direct and easy to understand. It

has a subject-first sentence structure making clear who did what,
uses full verbs,
avoids the semantically empty passive construction,
is shorter and more decisive in its language
removes unnecessary fluff.

Writing Abstracts and Introductions

Given that you win over or lose the review early on, a compelling abstract is vital for your paper. It should entice reviewers to read on and engage with your work. A good abstract clearly outlines the area, purpose, results, and significance of your research, and tells the reader what to expect (remember: spoilers are good).

Since abstracts are your calling card, an effective abstract should succinctly describe:

The general area
The predominant current approach
What is missing, or common problems with that approach (the “pain”)
What do you do instead (the “gain”)?
How do you do it?
Your results
What that means for the future

As a practical algorithm, start by writing one sentence for each of those points. It will not be the most beautiful abstract, but it will get your point across. Personally, I am a fan of writing abstracts early, even if you have not yet started the work (advice I got from a PhD committee member). It helps you clarify your thoughts, gives you a marching plan, and lets you gauge interest in the topic by sharing it with colleagues. Any research will reach a point where you need to decide which direction to take it in. Having a written statement of what you thought you would do in the form of an abstract helps you decide which path to go down.

That said, the abstract is the part of a paper that needs the most work, and that changes the most. It needs to accurately reflect the state of the work (which might have changed over time).

As an example of how to overhaul an abstract, take the following one from an old review (slightly altered for anonymity):

The available statistical methods for sequential tagging, especially in maximal NP (noun phrase) chunking, perform well. However, existing tools cannot combine methods freely. All the freely available tools consist of tightly coupled modules. Combining well-known methods (HMM, Maximum Entropy model, TnT, first and second order transition model, beam search, CRFs, smoothing techniques) in a general-purpose sequential tagger would help to find the best combination for each task. This paper introduces an updated, modular, universal sequential tagger which was used for maximal NP chunking. The tool was rewritten from an existing project and its Maximum Entropy based HMM (MEMM) framework was completed with other common methods: TnT-like trigram transition model and the ability of drop-in replacement unigram model from any classifier from the Scikit-learn library or compatible API. The new tool is open source under the LGPL licence. Several kinds of method combinations were tested for English and Amharic. A trigram based solution resulted in the best ever F-score of 93.56% for Amharic NP chunking (+3% improvement). The key of the improvement is based on more accurate POS categories with fine-tuned trigram search.

It has all the information needed, but dives right in, jumps around between parts, and buries some of the information. Using the 7-sentence structure and some of the basic rules we have seen, we can rearrange and rewrite it like this:

Noun phrase (NP) chunking is a core method in linguistic analysis.
The best NP chunking tools use statistical methods for sequential tagging.
However, existing tools cannot combine methods freely to find the best combination for each task.
We show how to combine well-known methods (different learners, models, and smoothing techniques) in a general-purpose sequential tagger.
We introduce a modular, universal NP chunker, based on Maximum Entropy Markov Models, which includes other common methods, like drop-in.
We test several combinations for English and Amharic and achieve the best reported F-score of 93.56% for Amharic NP chunking (+3% improvement).
The improvement is based on more accurate POS categories with fine-tuned trigram search.

A nice bonus of the 7-sentence abstract is that it also serves as a scaffold for your introduction section. Simply copy the abstract, expand each sentence into a paragraph by adding examples and references, and you have the outline of that section as well. Spend some time crafting a succinct example that encapsulates what you are working on. Make clear why it matters. Bonus points if you can reuse the example throughout the paper.

Grant Writing Essentials

At some point, you will find yourself in a position where you have to write a grant. Maybe it’s your choice, but it might also be a requirement to get tenure or a promotion. Either way, grantwriting has a couple of benefits independent of whether you end up getting the money. It sharpens your research ideas, forces you to think long-term, and helps you practice outreach to non-academics (many grant reviewers are not academics).

The mistake I made when writing my first grant (which, from what I’ve heard over the years, is common) was treating it as just a longer research paper. However, they are patently not. Writing grant proposals differs from writing academic papers in several ways. The What–So-What tradeoff is very different from a paper. It’s probably better to think of grants as more like blog posts: they should be understandable to a broad, self-selected audience, make a succinct point, and be somewhat exciting.

	Papers	Grants
Audience	Researcher with scholarly passion	Sponsor-centered: Service attitude
Focus	Past work you have done	Future work you wish to do
Function	Exposition: Explaining to reader	Persuasive: “Sell” the reader
Style	Impersonal, objective, dispassionate	Personal: Convey excitement
Organization	Individualistic: present your idea	Team-oriented: Feedback needed
Specificity	Specialized terminology ok	Accessible general language
Themes	Theses, theories, and ideas	Goals, activities, and outcomes

Grant Questions

Pick a problem you believe in and ask:

Why is it important?
How is existing knowledge or practice inadequate?
Why is your idea better?
How is it new, unique, different?
What will it contribute and who will benefit from it?

Also identify whether it’s an individual or consortium grant, and whether it’s top-down (they provide the topics) or bottom-up (you come up with an idea). Those aspects will matter in whether you start writing right away or find some collaborators first, and how you frame the idea, respectively.

Pitch Perfect

Make a compelling case for your idea’s impact and benefits. Ideally, the grant topic is:

an important need or issue that should be addressed
a gap between where we are now and where we could be
a limitation of current knowledge or way of doing things holding us back

All this is similar to the “pain” we have seen in Abstract writing. I have heard this called the “unrejectionable “: what major issue do the funders allow to continue by not funding your proposal?

In contrast, your grant project is an opportunity to:

advance our understanding or address a societal need
improve efficiency or lower costs of goods and/or services
reshape our thinking or way of doing things

And yes, that is the “gain” we have seen earlier.

With all that in mind, set the stage to engage and excite the reviewers. Remember: You win or lose on the first page!

Identify the need/importance: Who cares?
Describe your project’s fundamental purpose
Create a vision (“So What?”)
Summarize the state of the art
Show how your work will advance the field
State your solution clearly
Describe technical challenges to solving the problem
Describe the concept and establish credibility
Outline potential benefits
Envision the world with the problem solved

Keywords and Key Words

Clearly state the project’s goal and how it aligns with the grant’s objectives. They will give you specific keywords, sometimes explicitly: Pay attention to them and what they mean. Make sure you use those words (and maybe even boldface them) in your proposal.

Some other “grant” words:

Objectives: The big goals to solve. State them explicitly and refer back to them.
Work packages: Steps in your pipeline (e.g., data collection, model development, literature review).
Milestones: Achievements. Release of a model, data set, etc. These are often the results of work packages.
Deliverables: Overall outcomes of the project: e.g., a website, publications, models, data sets, workshops, etc.
Stakeholders: Society, academia, industry, and grant funders. Ideally, they all benefit from your findings in some way.

Money, Money, Money

Your grant officer should help you with the details, but you need to have the basics:

personnel (postdocs, PhDs, RAs, annotators)
teaching buyout/time costs
travel costs (flight, hotel, registration)
publication costs
equipment (GPU servers, eye trackers, etc.)
data or annotation services (student annotators, MTurk, surveys, etc.)

Account for all expenses, and make them make sense in the narrative. I.e., do not ask for something that is not justified by the project.

Universities will also take a fixed percentage of Overhead Costs: think of this as a fee for using their facilities and support. It varies by university, from 30% to over 60%. Some grants explicitly include it, others do not. Be sure to check!

Common Mistakes

Frequent (preventable) reasons for rejection include:

Proposal did not match program/keywords
Applicant did not follow directions
Problem statement is too global
Problem has no relationship to reality, or no potential solution
Problem has been studied to death already
Unclear problem, central theme, methods, or goals
Overly dramatic imagined consequences
Jargon and lack of clear writing

Beyond that, it’s of course up to the reviewers and funders whether they like your grant, but it’s better to avoid getting sorted out at an early stage already.

Conclusion

Both scientific papers and grant proposals should be clearly written to effectively communicate their potential impact and win over reviewers. The tools and strategies outlined here give you a scientific writing foundation to achieve that. Beyond that, aim for clarity, structure, and persuasiveness. Use concise language and tailor your message to your audience for the best results. But most of all: start early, keep iterating on the draft, get lots of feedback, and have fun with it. Happy writing!