Engineering

How to write a page an AI will quote

20 August 2026, 8 minute read

Much of the advice about answer engine optimisation repeats older search advice. I spent a day applying it to this site and measuring the result, and these are the parts that differ.

How do the two jobs differ?

A search engine ranks pages, while an answer engine extracts sentences. A page built to rank well can still give an answer engine nothing it can quote.

Ranking rewards a page that covers a topic thoroughly. Extraction rewards a page containing a self-contained sentence that answers a question without needing the paragraph around it.

Sentences that make sense on their own

The test: cut any sentence out of your page, show it to someone who has not read the rest, and see whether it still means anything.

Does not survive: "We founded it in 2025 and have been growing ever since." A reader cannot tell who "we" is or what was founded.

Survives: "Purcell Ventures LLC is a Georgia software company founded by Elijah Purcell in April 2025 that builds AI systems and software for small businesses."

The second sentence reads worse, and a model can use it. State each important fact once in that shape, usually near the top, and write the rest normally.

Third person for facts about a person

A page that says "I am nineteen and I run a software company" cannot be lifted and presented as a fact about a named entity, because a model has no way to attach "I" to anything.

So my site has two pages about me. The About page is in first person, which is how people want to read it. The answer page is in third person, so an engine can use it. They link to each other and say which is which.

FAQPage schema

FAQPage is the format that hands a machine a question and a self-contained answer as structured data.

The rule: the visible body of the page must contain every answer in the schema. Schema that promises an answer the page does not contain works like a misleading headline, and search engines can treat it as spam.

Write each answer so it makes sense read alone. Avoid "as mentioned above" and "our," and name the thing each time.

One ID for each person or company

If your site names a person or company in several places, every one of those structured data nodes needs the same @id. Otherwise you are describing several entities that happen to share a name, on your own domain.

Before I checked, my site had twelve distinct Person nodes, three of them anonymous. Now it has one.

{ "@type": "Person",
  "@id": "https://example.com/#founder",
  "name": "..." }

The same applies to prose. Pick one bio sentence and use it identically on every profile. If each platform has a slightly different bio, a model may treat them as different people with the same name. One plain sentence repeated exactly ties the profiles to one person.

How do you tell apart people with the same name?

If somebody shares your name, say which one you are. schema.org has disambiguatingDescription for exactly this, and the plain-prose version belongs on the page too.

Four other people share mine. My answer page names them and says I am none of them, so an assistant does not have to guess.

llms.txt

llms.txt is a plain-text file at the root of your site describing what is on it and where. Adoption is not universal and nobody should promise you results from it.

It takes twenty minutes and makes you write down what each of your pages is for, in one sentence, without marketing language. That exercise is useful by itself.

Inline source citations

Cite external sources in the text, and link them.

A 2026 survey of the field of generative engine optimisation research reports citation gains for pages that cite authoritative sources inline, with the biggest gains going to lower-ranked sites. The same survey reports that statistics, definitions, prices, dates and quotations improve citation, and that keyword stuffing reduces it.

That survey also found no established evidence that content volume, publishing speed or structured data are primary ways to improve citation. Much of the answer engine advice I read while doing this focused on those three things.

Indexing comes first

I did every technique above and then checked how many of my pages Google had indexed. It had seven of the twenty-eight I submitted. The structured data sat on pages Google had not indexed, and the cause was one line of inherited metadata telling Google to skip them.

So before optimising for extraction, confirm that Google has indexed your pages, because structured data on an unindexed page has no effect. I run nine checks for this.

Questions

What is answer engine optimisation?

Answer engine optimisation means writing a page so that an AI assistant can extract a correct answer from it. Search engine optimisation is about ranking the page in a list of results.

Part of the guide AI websites and getting found by AI search.