Writing is essential for two outcomes:

• allowing you to fully understand what you want to say
• allowing other people to understand what you want to say.

These may seem somewhat distinct but are intrinsically related: there is no way anybody else will *actually* understand what you intend to explain if *you* do not fully understand it.

The ultimate goal of writing is to remove all irrelevant information from an idea or notion, and structure the remaining information in a logical way such that the reader can follow *your line of thought* without any distraction from start to end.

Students often think that more is better (“How many pages should I write?”), somewhat assuming a positive correlation between relevance of an idea and number of words needed to explain it. If you think of writing in the terms denoted above, you'll often find the opposite is true.

Remember that a manuscript is not a report of the work you have done. It is a synthesis of what you have learned from it, in terms of the problem the manuscript investigates. Anyone reading your manuscript should be able to understand and redo what you did, and by doing so reproducing your results. What is not needed to understand and reproduce the results does not belong to the manuscript.

No manuscript is perfect, and experience helps a lot. But you gotta start from somewhere, and we hope the notes in this document can help you in the journey. Have you got feedback on these notes? Write l.allodi@tue.nl and/or n.zannone@tue.nl, we are happy to revise where helpful!

Disclaimer: here we frame it in the scientific context, but this is largely true for *factual writing* too (i.e. manuscripts whose intent is to convey facts).

Your reader is an adversary (often referred to as the proverbial “reviewer 2”): they will be skeptical of anything you claim or conclude. Any unsupported claim, unclear logical connection, loose association between problem and methodology has the potential to kill the whole manuscript. Text that is precise, to the point, without unnecessary “textual decorations” is much less attackable. Hence, it'll be much more credible.

Your reader is lazy: they will not want to do work on your behalf. If there is something to understand or conclude, or a link between concepts or observations, do not expect them to take on the effort to do it. They will not. You have to do it for them. Don't be lazy, so your reader can.

Manuscripts are typically structured in sections. These have the purpose of establishing the logical flow of the manuscript as a whole. Whereas these are fundamental and must make sense, the actual magic happens *within* the sections.

• The building block of a manuscript is the paragraph. Every single paragraph must have a precise purpose, a specific, single message it wants to convey or explain.
  • If removing the paragraph does not change the overall meaning of the section, trash it.
  • If the paragraph is trying to convey more than one message, you often need more than one paragraph.
• Every sentence in a paragraph must have a precise purpose and contain only the information needed to convey or explain it. For each sentence ask yourself:
  • Is this sentence fundamental to fully convey the message of the paragraph? If the answer is no, kill the sentence.
  • Which words in this sentence are not necessary to convey its message? Find them and rewrite/adjust the sentence accordingly.
• Figures and Tables are a great tool to better convey a message. They must however be presented in a way that they can be understood without reading any of the text. Use them as a tool to help the reader follow your thought process. You find some tips below under “Figures and tables”.

• Repetition is the death of writing. It's a symptom of weak ideas and/or bad structure. Think carefully if what you are writing overlaps with something you wrote earlier (or plan to write later). If it does, think of why and fix the issue.
• Write sentences as short as possible. Long ones are difficult to follow and lead to convoluted messages.
• Write in the present tense (remember: your manuscript is not a report of what you did!).

RQs give a natural structure to your research and clearly state what are the answers you are seeking.

• You can have as many as you want, but do not make them too “simplistic” or “narrow” (e.g. “How much time does it take our subjects to do X?”). Think more broadly about the relation of the question to the problem you investigate, and encode it in the question itself (e.g. “How does subject experience in Y affect the execution time of task X under conditions Z?”).
• Consider having a broader main research question, and a few sub-research questions. It often is a good way of conveying the broader ambition of the research (main RQ) as well as the breakdown of the problem leading to it (sub RQs).
• Wording in RQs is very (very, very) important. Think of them as a contract with your reader. Anything specified in there must be addressed as expected. Choose your words carefully and precisely.

• One figure/table, one message. If the message is not fundamental to understand the manuscript, consider moving it to an appendix.
• Figures/tables must be understandable by skimming through the manuscript. For this use clear labeling, colors where appropriate, and clear captioning.
• Each figure/table must be clearly referred to and fully described in the text.
• Captions should be precise and describe what the figure/table represents.
  • Captions of figures go below the figure.
  • Captions of tables go above the table.
• Little hack: you can consider adding a “minipage” in the LaTeX environment between figure/table and caption to explain it. This can be helpful when the figure/table is particularly complex.

• Citations are fundamental to support claims that are not directly supported by the data you are presenting. Never leave unsupported statements in the text.
• Avoid bulk citations whereby a sentence has several many citations supporting it. Choose instead the few ones that matter the most.

We refrain from giving a “go-to” structure for your manuscript. However a few writing guidelines can help clear up some confusion we often observe:

• Do not start writing from the section Introduction. That's the last one to write, as it contextualizes the whole work including results and contributions.
• “Background” and “related work” are not the same: the former presents key concepts needed to understand the present work (e.g. definitions, known structure or properties of the problem, ..); the latter presents other work that addresses the same problem, or problems close or related to the one you address.
• A well written Related Work section does not read as a list. (“[x] did this, [y] did that”). It should instead be a multi-faceted discussion of previous results about a theme: “regarding problem X, a few studies found that A is true[x,y]; by contrast, recent results find that A is likely not true under certain conditions [z]”.
• Make sure that you have a (sub)section where you explain the research/problem gap (i.e. what prior work have not yet answered about the problem you investigate). You can have research questions defined in there.
• There typically is a “Methodology” section where the manuscript explains the devised research framework and its implementation. The tighter the methodology is to the research questions, the clearer it generally is.
• The “Results” section is about reporting findings: numbers, quotes, experimental outcomes, etc.
• The “Discussion” section is about explaining why those findings matter.
• Use subsections to further structure a section in logical compartments. If you can avoid subsubsections, do it.
• Avoid having only one (sub)subsection in a (sub)section. You probably don't need it.

Manuscripts are about explaining *your* idea. An LLM has no clue what your (or any) idea is. As such it is a *terrible* tool to write content, and the manuscript will show – big time. It'll be generalist, approximate, contradictory, hallucinated; and your reader will notice.

LLMs can instead be a good tool to review text, or find ways of rewriting a sentence in a clearer, grammatically more correct, way. It is however tricky: sometimes the nuance or actual meaning of the sentence is lost in the re-write. Do NOT take for granted that the LLM output preserves the meaning of its input. Always review it critically.

LLMs can also be helpful to think about what other people may understand from your text. Sometimes asking an LLM to interpret and/or explain a paragraph you wrote can reveal hidden messages or interpretations that you did not foresee. Use with caution.