The exact same fact can get quoted from one page and ignored on another
A florist had two pages saying almost the same thing about delivery. An AI assistant quoted one every single time and never touched the other — and writing quality had nothing to do with it.
The short answer
Assistants quote pages that answer one question clearly, in a sentence that still makes sense once it's lifted out of the page entirely — the same test a caption has to pass with no photo attached. The five changes that matter most: lead with the answer instead of a scene-setting introduction, use the buyer's own question as the heading, name the subject in every important sentence instead of leaning on 'it' or 'our product,' keep facts identical everywhere they appear, and make sure the page can actually be fetched without a login, a redirect, or a wall in the way.
If you have ever compared two of your own pages and felt genuinely confused about why an assistant quoted one and ignored the other, you are not imagining it. A florist we spoke with had two pages that said almost the same thing. One was a paragraph on the homepage, written the way you would talk to someone standing at the counter. The other was a single line buried in a PDF delivery policy. When we ran the same question through an assistant, it quoted the homepage every time and never once touched the PDF, even though the PDF technically had the more complete answer.
Nothing about the writing quality explained the gap. The homepage sentence could be lifted out on its own and still make sense. The PDF sentence depended on three sentences before it, plus a document an assistant could not easily open in the first place. That is most of what this comes down to. Not better writing. Sentences that survive being pulled out of the page they live on — the way a caption has to survive being read with no photo attached, because that is essentially the job an assistant is doing to your page.
Laid side by side, the pattern is easier to see than any single example makes it look:
| Question | Written for a person browsing | Written to be quoted |
|---|---|---|
| Opening line | Sets the scene, then works toward the point | States the answer first, in one sentence |
| Heading | A category label, like Pricing | The buyer's actual question, like How much does it cost |
| Subject | Leans on it or our product once the topic is established | Names the product or business every time a fact is stated |
| Facts across pages | Close enough, written by whoever last updated that page | Identical everywhere, from one source of truth |
| Access | Fine for a person who clicks through a login or a wall | Fetchable with nothing standing between the crawler and the content |
Why does the first sentence matter so much?
Most business writing opens with context. A little history, a sentence about the team, then eventually the answer. That order makes sense for a person who arrived to browse. It works against you with an assistant, which is usually looking for the one sentence that answers the exact question it was asked, and has no patience for the three sentences before it.
Put the answer first. Save the context for after. A page that opens with “We deliver across the metro area within two days, with same day delivery available before noon” gives an assistant something it can lift whole. A page that opens with a story about how the business started gives it nothing to work with until three paragraphs later, if it keeps reading at all.
Should headings really be full questions?
Yes, and this is the cheapest change on the list. A heading like Pricing tells an assistant almost nothing about what it will find underneath. A heading like How much does a standard delivery cost tells it exactly which question the paragraph below is prepared to answer, in the buyer’s own words rather than yours.
This is also usually the fastest way to find gaps. Write out the ten questions your buyers actually ask, then check whether any heading on your site matches. Most businesses find two or three questions with no home at all.
Why does naming the subject matter more than it used to?
A sentence written for a human reader can lean on the paragraph around it. “It ships within two days” makes sense on a page where the previous sentence named the product. Pulled out on its own, an assistant has no idea what it refers to, so it either skips the sentence or has to go find the antecedent itself, which it will not always bother to do.
A sentence a model can quote is one that still makes sense with nothing around it.
Write the important sentences the way you would write a caption under a photo with no other text on the page. Name the product, the service, or the business every time a fact is stated, even when it reads slightly more formal than you would normally write.
What happens when the same fact says two things?
A price on the homepage that does not match the price on the order page. Hours listed one way in the footer and another way on a location page. A return window stated as thirty days in one place and a month in another. None of these are lies, exactly. They are just the kind of small drift that happens when a page gets updated and its twin does not.
A person skims past this. An assistant comparing sources treats it as a conflict, and a conflict usually means the fact gets hedged, attributed to someone else’s page instead, or left out of the answer altogether. Pick one place to be the source of truth for a fact, and update every other mention of it from there.
Can a technical problem undo all of this?
Yes, and it is the change teams are most likely to skip because it does not feel like a content problem. If a page sits behind a login, a cookie consent wall that has to be clicked through, or a redirect chain before the real content loads, the page may never be read at all. It does not matter how well the sentence underneath is written if nothing ever gets that far.
Does this apply to more than blog posts?
Yes, and it usually matters more on the pages that were never treated as content in the first place. A pricing page, a shipping policy, a specifications table, a support article — these get asked about constantly, because they answer exactly the kind of narrow question a buyer types into an assistant. The habit is the same one: find the sentence that answers the question, and make sure it can stand alone. A price sitting in a table cell with no sentence around it is invisible to a model in almost the same way a fact buried three sentences deep in a blog post is.
- 1
Answer first
The direct answer belongs in the first sentence under the heading, before any context or story.
- 2
Question as heading
Write the heading the way a buyer would ask it out loud, not as a category label.
- 3
Name the subject
Every important sentence should make sense with nothing else on the page around it.
- 4
One version of the truth
A fact stated in two places has to say the same thing in both, or it will be trusted in neither.
- 5
Confirm it can be fetched
Check the page loads with no login, no wall, and no redirect chain standing between it and the content.
These five changes cost very little to make and nothing to maintain once they are habit. The harder question, which questions to actually prioritise, is one we covered in the buyer questions that actually decide a sale. Once the pages are ready, the next thing worth checking is whether the assistant is naming you at all, which we get into in why your brand can be all over ChatGPT with your traffic unmoved.
Common questions
Why does an AI assistant quote one page and not another?+
Assistants tend to pull sentences that stand on their own. A sentence that answers a question directly, names its subject, and does not depend on the paragraph before it is far easier to lift into an answer than a sentence written as part of a longer story.
Does this mean I should stop writing naturally?+
No. It means the first sentence under a heading carries more weight than it used to. Write the rest of the page however suits your readers, and give the opening line of each section the same care you would give a caption.
How specific should the heading be?+
As specific as the question a buyer would actually type or ask aloud. A heading like Pricing invites a generic answer. A heading like How much does a two bedroom install cost invites the assistant to quote your page for that exact question.
What happens if the same fact appears differently in two places?+
The assistant has no way to know which version is current, so it may cite the wrong one, hedge the answer, or drop the fact entirely. Consistency is not a nice to have here. It is often the difference between being quoted and being skipped.
Can a technical problem stop a page from being quoted at all?+
Yes, and this is the change teams miss most often. If a page requires a login, sits behind a cookie wall, or redirects before the content loads, it may never be read in the first place. No amount of rewriting fixes a page that cannot be fetched.
Does this apply to more than blog posts?+
Yes, and it usually matters more on pages that were never treated as content in the first place — a pricing page, a shipping policy, a specifications table, a support article. These get asked about constantly, because they answer exactly the kind of narrow question a buyer types into an assistant. A price sitting alone in a table cell with no sentence around it is nearly as invisible to a model as a fact buried three sentences deep in a blog post.
What's the fastest way to find the gaps on my own site?+
Write out the ten questions your buyers actually ask, then check whether any heading on the site matches them word for word. Most businesses find two or three real buyer questions with no page, and no heading, actually answering them.
See which of your pages already read this way
Run a free snapshot and we will show you which buyer questions your site currently answers well, and which ones an assistant has nowhere to look.
