AI tools helped analyse the texts, gather the research and write the first draft of this article. I edited it and checked every quote against the original page. The cover illustration is AI-generated. Details at the end.
Read a typical explainer about a DeFi loan and count the words in each sentence. Often you get the same number again and again: twenty, twenty-two, nineteen, twenty-one. Every sentence has a clause, a comma and a qualification. Nothing is wrong with any single sentence, yet by the end the reader cannot say which one mattered.
Sentence rhythm is the pattern of long and short sentences in a paragraph. It is usually discussed as a matter of style, something that makes prose pleasant. This article argues that it is also a tool for explaining. A long sentence can hold a chain of conditions together; a short sentence can mark the point where the chain arrives. When every sentence is the same length, the reader has to find the conclusion alone.
This is the seventh part of a series on voice and trust in editorial work. Earlier parts covered showing evidence, disclosing AI use, correcting mistakes, attributing research, keeping confidence in proportion to evidence and keeping an authorial voice when you use AI. This one looks at a smaller unit: the sentence, and what its length tells the reader.
What three analytical writers actually do
To avoid relying on taste, we measured. Our style study took three writers whose work explains technology and finance to a general reader: Ethan Mollick (One Useful Thing, on AI), Benedict Evans (essays and newsletter on technology markets) and Molly White (Citation Needed, on crypto and fraud). The corpus contained 78 public texts and about 160,000 words written by the authors themselves: 24 texts by Mollick (2022–2026), 28 by Evans (2014–2026) and 26 by White (2022–2026). Sentence lengths were counted automatically by the same script for all three [1].
| Measure (average per text) | Mollick | Evans | White |
|---|---|---|---|
| Mean sentence length, words | 19.7 | 23.7 | 24.6 |
| Spread of lengths within a text (standard deviation) | 11.0 | 14.3 | 13.9 |
| Sentences of 35 words or more | 10% | 21% | 22% |
| Sentences of 6 words or fewer | 10% | 11% | 9% |
| Questions, share of sentences | 4.3% | 13.7% | 2.2% |

Three things stand out.
First, nobody writes in uniform sentences. The spread is large for all three writers. A standard deviation of 11 to 14 words around a mean of 20 to 25 means that a reader regularly moves between sentences of ten words and sentences of forty.
Second, long sentences are common. For Evans and White, more than one sentence in five is 35 words or longer. In the examples below, they are not careless run-ons; they carry a sequence: a cause and its consequence, or a piece of evidence and its qualification.
Third, the share of very short sentences is almost the same for all three: about one in ten. Their mean lengths differ, their subjects differ, their relationship with the reader differs, yet the short sentence appears at nearly the same rate. Our comparison concluded that the short accent sentence is a convention of the analytical newsletter genre rather than a personal trademark [1]. That makes it useful: it is a technique anyone can learn, not a voice to imitate.
How far these numbers go: three writers, one genre, automatic sentence splitting, and comparisons made between the three writers rather than against general English. They describe what these writers do. They do not prove that readers understand them better because of it.
Long sentence, short verdict
The clearest pattern in White's work is a long documented sentence followed by a short conclusion [2]. In her report from the fourth day of the FTX trial, she walks through code shown in court. One sentence explains, in 50 words, why the way a fund balance was being changed is hard to explain innocently. The next sentence has four words: "That would be fraud." [3]
The long sentence does the work of evidence. It keeps the conditions together so the reader can follow the reasoning without losing a step. The short sentence does the work of judgement. Because it is so much shorter than everything around it, the reader is unlikely to miss it or to mistake it for another qualification.
The same move appears in a very different text. In an essay about the early web, White builds an extended metaphor of gardens and plots across sentences of 44 and 36 words. Then: "But then, the walls went up." [4] Six words turn a description into a turning point.
Notice the order. The short sentence comes after the evidence, not before it. A short sentence placed first is a claim waiting for support. A short sentence placed after the support is a conclusion the reader has already been shown.
Short sentence as a plain-language reset
Mollick uses the short sentence differently. His mean sentence length is the shortest of the three and he uses short single sentences [5]; in our reading some of them follow a technical passage to say plainly what it means in practice.
In a 2023 guide to working with AI, one 47-word sentence explains which AI tool most readers would end up using and on what terms. The next sentence is four words: "It is also weird." [6] Then he explains what he means. The short sentence works as a reset. After the details, the reader gets the practical consequence in the simplest possible form, and the paragraph continues from there.
This is especially useful when the subject is technical. A long sentence about model versions, access and pricing is accurate but tiring. A short sentence afterwards tells the reader which part to keep.
Questions as steps
Evans has the most uneven rhythm in the study: the largest spread of sentence lengths, with long and short sentences both frequent [7]. His most distinctive tool is the question. Almost one sentence in seven is a question, three to six times the rate of the other two writers [1].
His questions are not rhetorical decoration. In a 2015 essay about autonomous cars, he works through second-order effects as a sequence of questions: how travel might shift from public transport, how many people might visit busy central areas they used to avoid, what that means for people on low incomes. The sequence ends with the shortest of them: "What does it do to cycling?" [8]
Each question opens the next step of the argument and tells the reader what the following sentences will try to answer. A run of questions also changes the rhythm: after long explanatory sentences, a short question resets attention without making a claim the author cannot support.
A wider view, with a caveat
The three-writer study is the measured part of our research. A broader survey of 148 writers included rough sentence counts for single texts by about fifteen of them, and those numbers are approximate because page navigation sometimes entered the counts [9]. Even so, the range was wide. In the texts we read, very short sentences made up about 19–30% of sentences for writers such as Morgan Housel and Warren Buffett, and only 2–5% for Vitalik Buterin, Ben Thompson and Jack Clark.
Housel's essay The Psychology of Money opens with a portrait built mostly from short sentences about a woman's quiet life, including "She never married. She never had kids." [10] Here the rhythm is the explanation: a run of short, flat facts shows how ordinary her life looked before the story reveals what was unusual about it.
So there is no correct ratio. Dense technical writers can work with very few short sentences; storytellers can work with many. What matters is whether the length of each sentence matches its job.
What style guides say
Official plain-language guidance mostly warns against long sentences. The GOV.UK style guide is blunt: "Do not use long sentences. Check sentences with more than 25 words to see if you can split them to make them clearer." [11]
The measurements above sit in some tension with that advice, though less than it first appears. The guide's practical test is a second look at anything over 25 words, and the three writers we measured regularly go past that mark when a sentence carries a sequence. In our reading, a long sentence that carries one ordered chain, such as a cause, a mechanism and a result, keeps visible the connections that fragments would leave the reader to rebuild; we have not tested this with readers. A long sentence that carries two unrelated ideas should be split. GOV.UK also writes for everyone who uses public services, which is a different job from an analytical newsletter's.
A worked example
The token, the loan and the protocol below are invented for illustration.
A reader has borrowed $1,200 in stablecoins against one token worth $2,000 on a lending protocol. The protocol allows the loan to reach 80% of the collateral's value. Here are two explanations of what happens next.
Version A, uniform rhythm. "The protocol allows a loan of up to 80% of the collateral's value before liquidation becomes possible for the borrower. Your loan of $1,200 against collateral worth $2,000 currently uses 60% of that value, which leaves you some room. If the token's price falls to about $1,500, the loan reaches the 80% limit set by the protocol's rules. At that point the protocol can sell part of your collateral to repay the debt, usually with an added penalty."
Every sentence is 19 or 20 words long. All the facts are there, but the sentence that matters most, the one that says you can lose collateral, looks exactly like the one about a ratio.
Version B, rhythm doing the work. "Your $1,200 loan is backed by a token worth $2,000, and the protocol lets the loan reach 80% of the collateral's value, so if the token falls to about $1,500 the loan hits that limit. Then the protocol can sell your collateral. How much? Usually enough to repay part of the debt, plus a penalty."
The first sentence is long, 35 words, because it carries one chain: loan, collateral, rule, trigger price. The second is short because it is the consequence. The question signals the next step, and the last sentence answers it. The facts are the same. The difference is that the reader can now see which sentence is the warning.

A method for your own drafts
- Find the conclusion sentence in each paragraph. If you cannot point to it, the paragraph may not have one. If you can, check whether it is shorter than the sentences around it. If it is the longest sentence in the paragraph, it is probably hidden.
- Let long sentences carry one chain. A long sentence works when it follows one order: time, cause or condition. If it changes subject halfway through, split it there.
- Put the short sentence after the evidence. A short verdict before the support reads as an assertion; after the support, it reads as a conclusion.
- Use one short accent per paragraph at most. Our own editorial rule is an average of 15–22 words per sentence, with no more than one short punch per paragraph [9]. Several short sentences in a row stop marking anything, unless, as in Housel's portrait, the run itself is the point.
- Use a question to open the next step. A question tells the reader what the next sentences will answer. It should be a question the text actually answers.
- Count. Write down the word count of each sentence in a paragraph. A row of near-identical numbers is a sign that nothing is being marked. This takes a minute and is harder to fool yourself with than reading for "flow".
Common mistakes and how to fix them
1. Uniform medium length. Every sentence twenty words, every sentence with a comma. Fix: find the conclusion and give it its own short sentence.
2. Staccato everywhere. Short sentence after short sentence, often in social posts. Nothing stands out because everything is emphasised. Fix: join the setup sentences into one sentence that carries the chain, and keep the short form for the verdict.
3. A short sentence the evidence does not support. A short verdict sounds certain. If the evidence above it is thin, the rhythm overstates the claim, the same problem an earlier part of this series discussed with a confident tone. Fix: make the claim match the evidence first, then decide its length.
4. Two ideas in one long sentence. The sentence is long because two thoughts were joined with "and", not because one chain needs room. Fix: split at the change of subject.
5. Rhythm instead of content. A well-timed short sentence cannot rescue a paragraph that has nothing to conclude. Fix: decide what the paragraph proves before you shape it.
What this article cannot tell you
The measured part of this article covers three writers in one genre, and the sentence counts come from automatic splitting, which makes small errors with abbreviations, lists and quotations. The study compared the three writers with each other, not with general English, and the coding of other features was done by rules and an AI agent rather than two independent human coders [1]. The rough counts from the wider survey cover single texts by about fifteen writers. We did not find, and did not run, a reader experiment showing that varied rhythm improves comprehension. What this article offers is an observed pattern in respected explanatory writing and a reasoned method, not a proven effect.
How this article was made
| Task | What the AI did | What a person did | How it was checked |
|---|---|---|---|
| Research | Collected the corpus of 78 public texts, counted sentence lengths and other features, wrote study notes for each writer | Chose the questions and the writers to study | Figures compared with the study's metrics file [1]; every quoted sentence found on the original page [3][4][6][8][10] and the GOV.UK guide [11]; word counts of the quoted passages recounted |
| Draft | Wrote the first draft from the study notes | Edited the text and the structure | A separate review compared figures and quotes with the sources |
| Images | Generated the cover illustration | Chose the illustration; designed the explanatory diagrams | Labelled as AI-generated in the disclosure line and alt text |
| Not done by AI | — | The decision to publish and responsibility for the text | — |
Conflicts of interest: none. The author has no relationship with the writers or publications discussed; the protocol in the example is fictional.
Responsible for this article: Dmitriy Lastovka, author. Corrections: contact form.
Sources
Labels: study (our own analysis), primary text (the writer's original page), guidance.
- Dapp.Expert style study, pilot of three writers: comparison and metrics (corpus of 78 texts, 160,115 authorial words; metrics file generated by the study's script, 4 November 2026). Study (internal, not peer-reviewed).
- Dapp.Expert style study, Molly White: rhythm and techniques (technique 8, short conclusion after a long paragraph). Study (internal).
- Molly White, "The FTX trial, day four: The fraud was in the code", Citation Needed, 9 October 2023. Primary text. https://www.citationneeded.news/the-fraud-was-in-the-code/
- Molly White, "We can have a different web", Citation Needed, 1 May 2024. Primary text. https://www.citationneeded.news/we-can-have-a-different-web/
- Dapp.Expert style study, Ethan Mollick: syntax and rhythm. Study (internal).
- Ethan Mollick, "On-boarding your AI Intern", One Useful Thing, 20 May 2023. Primary text. https://www.oneusefulthing.org/p/on-boarding-your-ai-intern
- Dapp.Expert style study, Benedict Evans: syntax, questions and rhythm. Study (internal).
- Benedict Evans, "Ways to think about cars", 22 August 2015 (page date; the URL path reads 2015/7/27). Primary text. https://www.ben-evans.com/benedictevans/2015/7/27/ways-to-think-about-cars
- Dapp.Expert style anatomy survey of 148 writers, section on rhythm and syntax (rough stylometry of single texts; house rule on sentence length). Study (internal).
- Morgan Housel, "The Psychology of Money", Collaborative Fund blog, 1 June 2018. Primary text. https://collabfund.com/blog/the-psychology-of-money/
- GOV.UK, A to Z style guide, "Sentence length". Guidance. https://guidance.publishing.service.gov.uk/writing-to-gov-uk-standards/style-guides/a-to-z-style-guide/










