Getting cited by answer engines

Search increasingly ends in a generated answer rather than a click. What decides whether your page is the one quoted — and how to measure it without guessing.

A growing share of commercial research now happens inside an assistant. The buyer asks a question, receives a synthesised answer with three or four citations, and clicks one of them — or none. Ranking fifth for a keyword is worth considerably less than being the source a model quotes, and the two are not optimised the same way.

This is not a new discipline so much as an old one with a different consumer. The reader is a retrieval system assembling an answer from fragments, and it rewards different properties than a human skimming a page.

What the retriever actually sees

An answer engine does not read your page. It reads a chunk of it — typically a few hundred tokens split at structural boundaries — scores that chunk against the question, and passes the winners to a model with an instruction to answer only from them. Three consequences follow directly, and they explain most of what works.

The crawler that does not run your JavaScript

# What a non-executing crawler sees.
curl -sL https://example.com/services | \
  python3 -c "import sys,html,re;t=sys.stdin.read();\
t=re.sub(r'<script.*?</script>','',t,flags=re.S);\
print(html.unescape(re.sub(r'<[^>]+>',' ',t))[:2000])"

Be an unambiguous entity

Models resolve entities before they answer. If your firm's name is rendered three different ways across your own site, the retrieval that should have found four supporting pages finds one. Pick the legal name and the trading name, decide the relationship between them, and state it identically everywhere — in Organization markup, in the footer, in the boilerplate paragraph, in the profiles you control off-site.

• One Organization block, in the root layout, with sameAs pointing at every profile you actually control.
• A short boilerplate paragraph — two sentences, the same words each time — on the pages describing the firm. Consistency is what makes an entity resolvable.
• Where a claim is checkable, attach the check: a named client, a named jurisdiction, a date. Assistants weight verifiable specifics heavily because they are what survives a fact-check.

The counter-intuitive part: publish what you cannot do

The pages of ours that get quoted most are not the capability pages. They are the ones stating a limit: that we do not hold ISO 27001 yet, that we do not publish rate cards and why, the list of situations in which a buyer should hire somebody else. Those passages are unusual, specific, and answer questions people actually ask an assistant — "is this firm certified", "what do they charge", "are they right for a small project".

There is a commercial argument as well as an honest one. A buyer who learns your limitation from a machine and calls anyway has pre-qualified themselves. A buyer who discovers it in week six of a procurement is a lost bid and a bad reference.

Measuring it without guessing

Rank tracking does not apply — there is no position to hold. Two measurements are worth running, one manual and one automatic.