On good writing

Documenting characteristics of good writing.
Aug 12, 2020Last updated: Sep 30, 2022

Crisp writing is an underrated skill. Good writing implies clear thinking, and clear thinking is invaluable. Here I collect resources that I keep visiting over and over.


  • It is always about the people.
  • "Something must be at stake".
  • Capture the larger story.
  • Write about at most one big thing.
  • Be relevant.
  • Be relatable.
  • Be conversational.
  • Make it happening.
  • Use metaphors!
  • Stick your landing, most interesting elements towards the end.
  • Simple words, clear ideas, and short sentences.
  • Don't shy away from complex ideas.
  • Do not overestimate what the reader knows.
  • Do not underestimate the reader's intelligence.


  • Use passive voice sparingly.
  • Narrative over lists.
  • Nouns not adjectives.
  • Cut mercilessly.
  • Callbacks.


General Communication

Mathematical Writing


Strunk's Rules

This is a summary from the excellent book by William Strunk Jr._, The Elements of Style.

  • [Rule 1] Always use 's (appostrophe s) to form singular possessive nouns. Some ancient proper nouns ending in -es or -is are an exception but the sentence can be restructured to sound better.

    Moses' laws can be rewritten as the law of Moses.

  • [Rule 2] With a series of comma-separated terms, no need for the last one.

    supervised, unsupervised, and reinforcement learning.

  • [Rule 3] Parenthetic expressions can be enclosed in commas. These must, however, be used sparingly. Often, one can compromise on succinctness and break such phrases into multiple independent statements.

    Linear regression, one of the most popular methods, is an example of supervised learning.

  • [Rule 4] In some sentences, connectors and or but do not explicitly define the relation between connected clauses. Instead, the sentences can be structured better using explicit relational qualifiers like although, while, etc.

    Linear regression is a regression method, but logistic regression is a classification method v/s While linear regression is a regression method, logistic regression is a classification method.

  • [Rule 5] A single compound sentence composed of complete clauses can be connected using a semi-colon ";" instead of a full stop; this is helpful to disambiguate pronouns like this, they, it, etc. Alternatively, we can fall back to using commas as usual with proper relational qualifiers.
  • [Rule 9] A paragraph should either denote a single unit in a composition or signal a new development of the subject from the previous paragraph. Single sentence paragraphs should be used sparingly, only as a connecting thread between the predecessor and successor.
  • [Rule 10] Within a paragraph, the first sentence can be the topic sentence, and the final sentence can emphasize a consequence of the discussion. Although, violating this thumb rule often breaks the monotony for the reader.
  • [Rule 11] Use active voice to enforce directness.
  • [Rule 12] Use positive statements; limit the use of not qualifiers.
  • [Rule 13] Omit needless words.

    She is the one who should be replaced by She.

  • [Rule 15] Express co-ordinate ideas in similar form. Correlative expressions should be followed by the same grammatical construction.

    the Bayesians and frequentists should be the Bayesians and the frequentists

  • [Rule 16] Keep related words together. Parenthetic clauses can be transferred to the beginning of a sentence to avoid reading interruptions.
  • [Rule 18] The proper place for emphatic words, which the writer desires to make most prominent, is towards the end of a sentence.

    Neural networks have hardly advanced our understanding of the human brain, though they are behind many breakthroughs in computer vision. v/s Neural networks are behind many breakthroughs in computer vision, but they have hardly advanced our understanding of the human brain.

© 2023 Sanyam Kapoor