# Citeability and share-of-model — getting quoted by AI, not just found

*2026-06-07 · Citeability & Answer Architecture*

> Search rewarded ranking; AI rewards being quoted. Here's how to make your content citeable — and why "share-of-model" is the metric that replaces the click.

## Found is not the same as quoted

For twenty years the goal was to be *found*: rank on page one, win the click. The
AI web introduces a different goal — to be *quoted*. When an assistant answers a
question, it assembles a response from sources it can read, trust, and lift cleanly.
You can be perfectly indexed and still never make it into that answer, because your
content wasn't shaped to be extracted.

Citeability is the discipline of shaping it so it is. It's the heart of the
[Citeability & Answer Architecture pillar](/framework#citeability), and it's where
generative engine optimisation (GEO) departs from classic SEO.

## Share-of-model: the metric that replaces rank

SEO gave us a clean KPI: where do you rank? GEO needs a new one, because there is
often no ranked list and frequently no click at all. The metric is **share-of-model**:
across the questions that matter to your business, how often — and how accurately —
do AI assistants cite *you*?

It's the AI-era equivalent of share-of-voice. You measure it by asking the
assistants directly: pose the twenty questions a prospect would ask, across the major
engines, and record who gets named, what they're told, and whether it's right. Track
that over time and you have a defensible picture of your presence in the answer layer.
(That measurement loop is its own discipline — see
[AI-representation observability](/guides/ai-representation-observability).)

## The four levers of citeability

Citeability isn't a writing style; it's four concrete, checkable properties.

### 1. Extractable answer units

A language model wants a bounded, self-contained chunk it can lift without dragging in
your nav, caveats, and cross-references. Lead every key page with an 80–120 word answer
unit that resolves the question on its own. This is the most important lever, and it has
its own guide: [answer-layer content](/guides/answer-layer-content). The box at the top
of this page is an example — copy it anywhere and it still makes sense.

### 2. Anchor-level addressability

Engines increasingly cite not just a page but a *passage*. If your claims live behind
stable fragment anchors (`#share-of-model`), an assistant can deep-link the exact
sentence it's quoting — and a reader (or another model) can verify it in place. Give
every meaningful heading and claim a persistent ID, and never let those URLs churn.

### 3. Canonical claims with an "as of" date

Models reward content that states facts plainly and says *when* they were true. A
sentence like *"As of June 2026, the European Accessibility Act applies to most
commercial websites"* is far more quotable than the same fact buried in a hedged
paragraph. Pair canonical statements with `datePublished`/`dateModified` and a visible
last-reviewed date so the engine can trust the recency.

### 4. Stable, durable URLs

A citation is only as good as the link behind it. Every URL change breaks accumulated
citeability and orphans the references models have already learned. Treat URLs as
permanent contracts.

## Why this is genuinely different from SEO

SEO optimised pages to be *ranked* against each other; GEO optimises passages to be
*lifted* into an answer. The unit shrinks from the page to the claim. Keyword density
stops mattering; clean extractability starts mattering. And crucially, the winner isn't
always the highest-ranked page — it's the most *quotable* one, the source whose wording
the model can reuse with least risk.

## The zero-click paradox

Here's the uncomfortable part. Do this well and you may see *fewer* direct visits, not
more — the assistant delivered your answer without sending the click. That's not failure;
it's the new shape of influence. Your content became the source the answer was built
from. The strategic response is to optimise for *being that source*, and to build the
relationship somewhere the assistant can't disintermediate: your email list, your
[agent surface](/framework#agent), your branded entity. Presence in the answer layer is
the asset; the click is just one way it used to show up.

## What to do this week

1. List the 20 questions a prospect would ask an AI assistant in your category.
2. Ask those questions across the major assistants today and record your **share-of-model** — who gets cited, and whether it's accurate. This is your baseline.
3. On your highest-value pages, add a self-contained answer unit, stable heading anchors, and a visible "as of" date to your key claims.
4. Re-run the questions in a month and watch the share move.