Answer-First Writing: Putting the Conclusion in the First Sentence

Answer-first writing means putting the conclusion in the first sentence of every section and earning it in the sentences after. Not at the end of the article. At the top of each part of it.

It feels wrong to almost everyone trained to construct an argument, because it gives away the ending. It is also how news has been written for more than a century, how military briefings are structured, and the single highest-yield editing change available for AI retrieval. You can apply it to pages you already have in an afternoon, which is rare for anything in this field.

The mechanical reason it works

Retrieval does not score your page. It scores chunks of it, usually a few hundred tokens, cut on headings and paragraph boundaries. That chunk is what gets embedded, compared against a question, reranked and eventually handed to a model. Your page is a container.

So the question that decides everything is whether a given paragraph still makes sense with everything around it removed. A paragraph opening This means that costs drop by roughly a third is worthless in isolation, because this pointed at something in the paragraph above and that paragraph is not coming along. The full pipeline is in how retrieval decides which sources to pull, and this is the property that decides survival at the first gate.

There is a second effect at the reranking stage that matters more and is less discussed. A reranker reads the question and the passage together and scores how completely the passage answers it. A paragraph that answers the question in its first sentence scores well. A paragraph that answers it in the last sentence, after context, scores worse, because much of what the reranker read was setup.

This is not a new idea and that is the point

The structure is the inverted pyramid, and it has been the standard for news writing since the telegraph made transmission unreliable enough that editors wanted the important part sent first. It survived because it turned out to match how people read.

Usability research reached the same conclusion independently. Nielsen Norman Group’s work on the inverted pyramid on the web found that readers scan, abandon early and rarely reach the bottom of anything, which makes a conclusion at the end a conclusion most people never see. Government writing guidance reached it too. The US federal guidance on organising information is explicit that the most important material goes first and that the reader should not have to assemble your point from scattered parts.

Academic writing guidance says the same thing about paragraphs specifically. Both the Purdue writing lab and the Wisconsin writing handbook teach that a paragraph opens with a sentence stating what it will establish. That is answer-first writing under an older name, and it means most of your team already learned this and then unlearned it for marketing copy.

If you want to know whether this is your binding constraint before rewriting anything, ask an engine a question you should win and look at which paragraph of which page it actually quoted. It is almost never the best-written one. Our free AI visibility check does that at a proper sample and shows you the exact passages winning against yours.

What it looks like, line by line

Here is the same information written both ways. Neither is badly written. Only one survives being cut out of its page.

BUILDS TO THE POINT  (loses at retrieval)

  Over the past few years the way buyers research software has
  changed considerably. Where they once relied on search results
  and peer recommendations, they increasingly begin with an AI
  assistant. This shift has consequences for how vendors are
  discovered. As a result, being absent from AI answers now costs
  more than being absent from page two of Google.

ANSWER FIRST  (survives)

  Being absent from AI answers now costs more than being absent
  from page two of Google. Buyers increasingly start their
  research with an assistant rather than a search results page,
  so a vendor that is never named is never shortlisted. The
  change happened over roughly three years and is most advanced
  in technical categories.

The second version says the same things. What changed is order and dependency. Every sentence in it can be understood without the one before it. The first version has three sentences that cannot.

Opening pattern	Why it fails a chunker	Replace with
This means that…	The referent is in the previous paragraph, which is gone	Restate the subject explicitly
As a result, …	Result of what? Nothing in this chunk	Name the cause in the same sentence
However, …	Contradicts something the model cannot see	State the claim, then the qualification
In other words, …	The other words were the ones that mattered	Delete the original and keep only this
It is also worth considering…	Says nothing before the comma and hedges after it	Say the thing

Where to apply it, and where not to

Not everywhere. Applied to every sentence it produces flat, hectoring prose that nobody enjoys, and enjoyment still matters because humans are also reading.

Apply it at three specific places.

• The first paragraph under every H2. This is where a chunk usually starts, so it is where the self-contained claim needs to be. Everything after it can flow normally.
• The opening of the page. The first two sentences should answer the question in the title plainly, before any framing. If somebody read only those two sentences they should have the answer.
• Any paragraph containing a number. Figures are the most quoted thing on any page and the most often stripped of their meaning. The number, the unit, the period and the subject belong in one sentence.

Between those points, write normally. Vary sentence length, use connectives, let arguments develop. A page written entirely in declarative openers reads like an instruction manual and gets abandoned by humans, which eventually costs you more than a retrieval slot.

The objection worth taking seriously

Marketers push back on this for a real reason. Answer-first writing removes suspense, and some persuasion depends on sequencing, on building a case before delivering a conclusion the reader would have rejected cold.

That is true for a sales letter and largely false for everything else on a company website. The buyer reading your documentation, pricing page or technical explainer is not being persuaded through narrative. They are looking for a fact and deciding whether you are worth more time. Withholding the fact to preserve a structure loses more of those readers than it converts.

The compromise we use is to keep narrative at the level of the page and answer-first at the level of the section. The article as a whole can still take the reader somewhere. Each section still opens by saying what it is about. That satisfies the chunker and the skimmer without producing something unreadable.

A thirty minute pass on an existing page

You do not need a rewrite. You need an editing pass, and here is the order that finds the most in the least time.

1. Search the page for this, it, they and as a result at the start of paragraphs. Every hit is a candidate failure. Replace the pronoun with the noun it refers to.
2. Read the first sentence under each H2 in isolation. If it does not make a claim, move the claim up from wherever it currently sits.
3. Find every number on the page. Confirm each one sits in a sentence that also contains what it measures, the unit, and the period it covers.
4. Delete transition-only sentences. Anything whose whole job is to connect two paragraphs is dead weight in a chunk and usually dead weight in the read.
5. Reread the opening two sentences. If they frame rather than answer, swap them for the answer.

Do this to your five most commercially important pages before commissioning anything new. It costs half a day and it addresses the two stages of retrieval you can actually influence. Pricing pages benefit most, for reasons covered in why your pricing page is the page AI reads most, and definition-heavy pages benefit almost as much, which is most of the argument for keeping a glossary.

How you know it worked

Watch for a specific change rather than a general one. You are looking for your own sentences coming back in engine answers, close to verbatim. That is the signal that a passage of yours is being retrieved and used rather than your page being vaguely known.

Expect four to eight weeks, gated by crawl frequency rather than by anything you control. If nothing moves in ten weeks and crawlers are definitely reaching you, the problem is upstream of writing. Either the pages are not being retrieved at all, or what the model already believes about you is overriding what it reads, which is the distinction in training data versus live retrieval.

---

Related reading

• The AEO technical playbook. Where editing sits relative to the technical work, in execution order.
• Schema markup for AEO. The other half of making a page machine-legible, and which parts of it are decoration.
• Your about page is an entity document. Answer-first applied to the page most companies write worst.
• What AEO is. The wider picture if this is your first piece of ours.