Put the claim of a paragraph in its first sentence and spend the rest of the paragraph supporting it. Most readers of anything professional or academic are skimming, and skimming in English means reading first sentences, so a paragraph that arrives at its point in the last line has put the point where nobody was looking.
The test takes a minute and it is brutal. Read only the first sentence of every paragraph in something you wrote, in order, and see whether you have the document. If you have a list of subjects rather than a set of claims, the paragraphs are announcing what they are about instead of saying anything.
Eight rules for building one
- one idea per paragraph. If you cannot say what a paragraph is about in one clause, it is two paragraphs.
- the topic sentence goes first, stating the claim, so a reader skimming the first lines gets the argument.
- then evidence, then the consequence. Claim, support, so what. This shape covers most expository paragraphs in English.
- three to six sentences is typical, and consistently longer paragraphs are usually two ideas that have not been separated.
- a paragraph of one sentence is allowed and is emphatic, so spend it on the thing you most want remembered.
- however, therefore and for example signal the turn, telling the reader what relationship the next sentence has to the last.
- end on the point, not on a summary. A paragraph that restates its own first line has wasted its last.
- a new speaker starts a new paragraph in dialogue, without exception, which is a rule of layout rather than of grammar.
A claim, not a topic
The difference is whether somebody could disagree with the sentence. This section discusses the costs of the delay is a label; nobody can hold the opposite of it. The delay cost more than the change would have is a claim, and a reader now knows what the paragraph is for and can decide whether to read the rest.
Labels feel safer, which is why they are everywhere. Stating the claim first commits you before the evidence has been presented, and that feels premature. In English it is what a reader expects, and withholding it does not create suspense, it creates the impression that the writer has not decided anything.
Where it stops: building to the point
The stall is a habit rather than an error, and it comes from a different writing tradition rather than from weak English. In several of the world's education systems a good paragraph, and a good essay, sets out the considerations and arrives at the position at the end, and a claim stated up front reads as arrogant or as unearned.
A writer trained that way produces paragraphs that are well organised, fully evidenced, and read as though nothing is being asserted. The feedback that comes back is vague: unclear, hard to follow, what is your point. None of it names the structure, so the writer adds more evidence, which makes it worse, and the loop can run for years.
Naming it is most of the fix, because it is a convention rather than a skill. Once a writer knows English readers arrive expecting the claim, moving the last sentence of each paragraph to the front is an afternoon's work on a whole document.
One idea, and the handoff
A paragraph holds one claim. When a second one arrives, that is a new paragraph, and the length has nothing to do with it: a two-sentence paragraph carrying one claim is fine and a page-long one carrying three is not.
The last sentence has a job too, and it is not to summarise. It hands off: it ends on the thing the next paragraph will start from, so the reader crosses the gap carrying something. A paragraph that closes by restating its own first sentence has spent its last line on the reader's behalf and given them nothing to take forward.
Asking something to find the claim it cannot find
Paste a draft into a model and ask it to quote, verbatim, the sentence in each paragraph that states that paragraph's claim. The instruction to quote is the whole of the technique: a model asked to summarise will helpfully invent a claim for a paragraph that has none, because charitable reading is exactly what it is good at, and a paragraph where it cannot produce a quotation is a paragraph with nothing in it to quote.
It replaced asking a colleague to read a draft, who also reads charitably, supplies the structure you left out, and reports that it was clear. Free at the level this needs. The limitation is that it will tell you which paragraphs lack a claim and cannot tell you whether the claims are any good, so it finds the structural fault and never the argument.
What it looks like when it goes wrong
The recognisable failure is a document that is completely clear to anybody who reads every word of it and incomprehensible to somebody skimming, which describes almost every reader who matters.
The second is a run of paragraphs joined by heavy adverbs at the front. If the connections between your paragraphs are being carried by words like in addition, the paragraphs are not connected, and the adverb is doing a job the content should have done: the previous paragraph should have ended on the thing this one begins with.
What holds a paragraph together at the level below this, and where a long sentence should be broken, is length, rhythm, and where a sentence breaks, and the place where a stated claim is not merely preferred but marked is thesis, argument and evidence, in the shape examiners expect. Both sit under emails, essays and reports that read like a native wrote them.
The open question is what happens when there is only one paragraph and somebody has to act on it. A request has a claim too, and burying it has an immediate cost rather than a slow one, because the reader answers the first thing they see: openings, closings and the register each situation asks for.