Answer engine optimisation from scratch

Search did not stop working. It grew a layer on top. A question that used to return ten links now returns a written answer with three or four sources attached, and the click that used to be yours goes to whoever got cited.

That is the whole problem. This guide covers what actually drives citation, in the order worth fixing.

How a generated answer picks sources

Every engine differs in detail, but the shape is consistent. The engine retrieves a candidate set, usually from its own index plus a live search; it ranks those candidates for relevance and trustworthiness; then it composes an answer and attributes the claims it used.

Two things follow from that.

First, ranking still matters enormously. Retrieval leans on the same signals classic search uses. A page that cannot be found is never a candidate.

Second, ranking is not sufficient. Being retrieved gets you into the candidate set. Being *quotable* gets you cited.

What quotable means in practice

A model composing an answer needs a self-contained claim it can lift without inheriting ambiguity. That means:

• Answer the question in the first two sentences. Content that builds to its point over six paragraphs is hard to quote and easy to skip.
• Make each claim stand alone. A sentence that depends on the previous three for meaning cannot survive extraction.
• Attach specifics. Numbers, dates and named methods are quotable in a way that adjectives are not.
• Structure with real headings. Headings phrased as the questions people ask give the retriever a clean match.

Entity coverage

Engines reason about entities, not strings. If your page discusses a topic without naming the concepts, products and standards that surround it, it reads as thin even when the prose is good.

The practical test: list the ten concepts an expert would expect in an answer on this topic. Count how many your page names. Most underperforming pages cover three or four.

Schema, honestly

Structured data helps, but it is oversold in this context. It clarifies what a page *is* — an article, a product, an FAQ — which assists retrieval and eligibility for rich results. It does not make a weak page quotable.

Add it because it is cheap and correct. Do not expect it to carry a page that does not answer the question.

Measuring any of this

You cannot manage citation share without measuring it, and it is not in your analytics. The approach that works:

1. Build a prompt panel — the questions real buyers ask, phrased as they would phrase them, not as keywords.
2. Run that panel against each engine on a fixed schedule.
3. Record which domains are cited, in what position, and with what framing.
4. Track the share over time and alert on displacement.

A hundred prompts sampled weekly tells you more than ten thousand sampled once.

The order to work in

• Fix retrieval first. If you do not rank, nothing downstream matters.
• Restructure your top twenty pages for direct answers.
• Close entity gaps on those same pages.
• Add schema.
• Then measure, and let the measurement drive the next twenty.

Teams that invert this — starting with schema and prompt panels while their pages still bury the answer — spend a quarter and move nothing.