Long-Form Answers for AI Summaries: What Gets Quoted

An assistant quoting your page is not reading it. It is lifting one paragraph. Write that paragraph on purpose.

Last updated: November 1, 2026

Somebody asks an assistant how a deposit works on a short stay, and the answer comes back in three sentences with your page as the source. What got used wasn't the page. It was one paragraph inside it that made sense on its own. Write that paragraph deliberately and the rest of the page stops having to carry the whole job.

Key Takeaways

• A summariser keeps passages, not pages. It scores chunks against a question and compresses whatever survives. Self-sufficiency is the whole game.
• The heading is a retrieval key. Written as the guest's question, it matches the words someone actually typed. Written as a category, it matches nothing.
• Put the answer in the first sentence. With the qualifier attached, so a lifted sentence isn't a false one.
• One paragraph that stands alone. No pronouns pointing backwards, the subject named again, forty to eighty words.
• Format buys you liftability, not inclusion. Whether an assistant quotes you depends on things you don't control, and no amount of structure changes that.

What a summariser actually lifts

A summariser doesn't read a page the way a person does. It splits the text into chunks, scores those chunks against a question, and compresses the ones that score. The chunk that wins is the one that answers the question while standing on its own two feet.

That single fact explains most of what goes wrong. An answer spread across four paragraphs loses to an answer that sits in one. A paragraph that opens with three sentences of background loses to one that opens with the conclusion. And a paragraph that says "this is why it matters" loses to one that names the thing again, because the summariser can't see what "this" refers to once the passage is on its own.

Here's the test worth running on your own page. Cover everything except one paragraph, then read it. Does it still answer the question it sits under? If it needs the paragraph above to make sense, it's not liftable, and no amount of good writing elsewhere will fix that.

Notice what the test doesn't reward. Clever openings. Narrative build. The slow reveal of the point in the fourth paragraph. All of that works on a reader who's committed to the page, and none of it survives something that reads one chunk at a time.

One example of the difference. A page can spend three paragraphs explaining how deposits work and never state the rule in a sentence, and it'll read as thorough to a human and as empty to a summariser. The same page with a two-sentence answer at the top of the section reads as blunt to a human and as usable to a machine. Both versions are honest. Only one of them travels.

The question-shaped heading

Write the heading as the question a guest would actually ask. Not "Deposit considerations". "How much is the deposit, and when do I get it back?"

The difference is mechanical as well as stylistic. A heading is a retrieval key: it's the strongest signal a summariser has about what the passage below it answers. When the heading uses the guest's words, the passage matches the question. When it uses your words, it matches a category nobody typed.

So take the vocabulary from the questions you actually get. If guests say "bond" in one market and "deposit" in another, use one in the heading and the other once in the paragraph. That's not padding; it's the difference between matching a query and almost matching it.

Three habits make headings work harder.

One question per heading. "Deposits and cancellation" is two questions in a coat, and a summariser will lift the answer to one of them and attach it to the other.

No cute headings. "The money bit" reads well to a returning visitor and maps to nothing. Charm costs you the match, and the match is the point.

No heading that promises more than the paragraph delivers. A heading that says "everything you need to know" gets a passage that answers one thing, and the gap reads as a bad source.

First sentence carries the answer

The first sentence under a heading should be the answer, complete, with its conditions inside it.

Compare. "Deposits vary." That's a sentence with no content, and it can't be lifted into an answer. "A deposit is taken at booking and returned within the period your platform's terms allow after checkout, provided nothing is claimed against it." That's an answer, and it survives being quoted on its own.

The qualifier matters as much as the answer. A blunt sentence is easier to lift, and a blunt sentence that's only true in some cases is worse than useless — it becomes a quote you'd have to correct later. Keep the condition in the same sentence as the claim, and the quote stays honest even when the surrounding page is gone.

Then put the background after. Why the practice exists, how it varies by market, what changed last year: all fine, and all of it belongs below the answer rather than above it. A reader who wants the detail gets it. A summariser that wants one sentence gets the one it needs.

One more small thing, and it's the one people skip. Say what the answer applies to. "In most European cities" or "for stays of seven nights or more" costs six words and prevents a quote from being read as a universal rule.

One paragraph that stands alone

The paragraph under the heading is the unit that gets lifted, so build it as a closed box.

Close it with three rules. Name the subject again instead of using "it" or "this". Include the scope — who, where, how long. Keep it between forty and eighty words, long enough to answer and short enough to lift whole.

Then date anything that moves. A paragraph about a rule, a fee or a platform setting without a date is a paragraph that will be quoted two years after it stopped being true, and the quote will have your name on it.

Here's where the editorial work pays twice. The paragraph built this way is also the paragraph a skimming human reads, because a person scanning a page on a phone behaves a lot like a chunk scorer: they land, they read one block, they leave. Writing for the lift and writing for the skim turn out to be the same craft, which is the reason to bother.

And that's the same shape as the questions you'd ask an assistant about your own operation. "Who's arriving today, who's in house, which rooms are due out" is a question with a short factual answer. It's also the kind of thing localsbnb.com answers from your own data when Claude, ChatGPT or Cursor is connected to the property: one question, no second app. The answers are read-only, covering arrivals, in-house guests, departures, room status, the room calendar, channel rates and the occupancy, ADR and RevPAR figures. The page above is writing. That one is a lookup.

FAQ

Does this mean I'm writing for machines now?

No. It means writing the paragraph that answers the question, then putting the rest of the page around it. A guest on a phone reads one block and leaves, and a summariser lifts one block and leaves. The same paragraph serves both, which is why the effort isn't wasted even if no assistant ever quotes you.

How long should the answer paragraph be?

Forty to eighty words. Under forty, there's usually a condition missing and the quote turns out wrong. Over eighty, the paragraph starts carrying background, and the part that answers the question has to be extracted before it can be used.

Will an assistant definitely quote my page if I write it this way?

No, and nobody can promise that. Format decides whether a passage can be lifted; retrieval, freshness and the assistant's own choices decide whether it's ever looked at. What you control is the shape of the paragraph, so make the shape right and leave the rest alone.

A page is a container. The paragraph is the thing that travels. localsbnb.com is where the questions about your own nights get answered from the calendar itself, one read-only question at a time, while the writing stays the part you do by hand.

---

This article is general guidance for hosts and isn't legal, tax or insurance advice. What a platform or an assistant does with published content changes without notice, and no page structure guarantees inclusion in any answer. Confirm the current terms of each platform you use and any local requirement that applies to your unit before you rely on published guidance.