How to be quotable — agentsite

# How to be quotable

Layer 4 of the AEO model decides whether the agent that reached your page actually quotes it. Four moves separate quotable pages from readable ones — answer first, cite specifically, link out generously, link in cohesively.

By AgentSite · 5 min read · Updated 2026-05-25

The first three layers of AEO decide whether an engine can reach your page, navigate your site, and read what kind of thing the page is. Layer 4 decides whether the page actually gets cited. Four moves separate quotable pages from merely readable ones. Most sites are missing at least three of them.

## The four moves

**Answer first.** The agent that reaches your page is looking for an extractable answer to whatever question the user asked. A 40–60 word paragraph at the top of the body, written as a self-contained answer to the implied question of the title, is the unit that gets quoted. Pages that bury the answer under introductions, marketing copy, or table-of-contents links lose the citation to pages that don't. The pattern lives in [direct answer](/direct-answer).

**Cite specifically.** Every load-bearing claim — every number, every "studies show," every "as of last year" — should name its source inline with a link. The Princeton GEO paper measured "Cite Sources" as a top-three optimization tactic with up to +40% visibility lift in controlled experiments ([Aggarwal et al., KDD 2024](https://arxiv.org/abs/2311.09735)). Vague attribution gives you the prose-cost of citation with none of the lift. The discipline lives in [statistics and citations](/statistics-citations); the broader category in [outbound citations](/outbound-citations).

**Link out generously.** Every external link in your body is a credibility transfer from the source domain to yours. A page that grounds itself in primary sources reads as substance; a page that asserts reads as opinion. Google's E-E-A-T framework names "clear sourcing" as a trust signal ([Google Search Central, creating helpful content](https://developers.google.com/search/docs/fundamentals/creating-helpful-content)). Three contextual outbound citations is the working floor for a claim-dense page.

**Link in cohesively.** Internal links are the authority topology engines read alongside the content. A page densely embedded in your site's link graph reads as central; an orphan page reads as peripheral. The discipline is in [internal links](/internal-links). The combination of inbound and outbound links is what makes the page a node in a graph instead of a standalone block — and engines built on retrieval-augmented generation are graph-aware by construction.

## Why Layer 4 compounds

The lower layers are infrastructure. Once you've fixed them, fixing them again does nothing. A site can't be more rendered than rendered, more reachable than reachable, more declaratively typed than declaratively typed. The work on Layers 1–3 is one-time architectural cleanup.

Layer 4 is editorial. It compounds. Every new page is another opportunity to apply the four moves — and every page that applies them adds to the corpus that answer engines learn to cite from your domain. Vercel measured 569 million GPTBot fetches and 370 million Claude fetches in a single month ([Vercel, Dec 2024](https://vercel.com/blog/the-rise-of-the-ai-crawler)). The bots are reading. The question is whether they find prose worth quoting when they read.

The asymmetry is sharp. Two pages on the same topic, both rendered cleanly, both with the right schema. One opens with a 50-word direct answer that grounds each claim in a linked source. The other opens with a marketing intro and footnotes its sources in a sidebar nobody clicks. Engines will quote the first and skip the second every time. The infrastructure is identical; the citation outcomes diverge.

## What the moves look like together

A page that nails Layer 4 has a particular shape:

- An H1 that names the topic.
- A first paragraph that answers the implied question of the H1 in 40–60 words, no setup.
- Each subsequent paragraph hangs claims off named, linked sources — not "studies show" but "[Vercel measured…](https://vercel.com/blog/the-rise-of-the-ai-crawler)."
- Three or more inline links to other pages on your domain, each one in body prose, none in the footer.
- A footer that gives the reader one obvious next page.

This shape is what `aeo.md` does. This page does it too — count the inline links. The point isn't to be clever; it's that the shape is mechanical, replicable, and easy to verify. Every published page in this corpus passes a build-time gate scanner that checks for the lede word count, the citation count, the internal-link count, and the body-length band. Layer 4 is one of the few places in software where "follow the checklist" is the right strategy.

## What sites get wrong

Three failure modes show up reliably across the audits we run:

1. **Buried answer.** The first paragraph is a setup, not an answer. The body has the right content but the lede is throat-clearing. The agent extracting the top of the body finds intro, not substance.
2. **Vague attribution.** The page cites "studies" and "experts" without naming them. The Princeton GEO paper measured this directly: pages with named, linked sources outperform pages with unnamed ones by margins that justify the editorial effort.
3. **Orphan pages.** The page has the answer, the citations, the schema, the rendering — and zero inbound internal links from anywhere else on the site. The link graph reads it as peripheral and engines weight it down accordingly.

The full catalog of failure modes by layer is in [AEO problems](/aeo-problems). Layer 4 specifically has its own concentrated cluster — buried answer, vague attribution, date inflation — all of which respond to the same fix: write to be quoted, not to be read.

## Where this fits

This is the body-side companion to [the five layers of AEO](/five-layer-aeo). The five-layer model names the structural surfaces; this page names what to do on the one surface that matters most after infrastructure is sound. The longer thesis on why citation is its own optimization target sits in [agent readability](/agent-readability).

If you want to see which of the four moves your site is missing on which pages, [run your AEO score](/score) — the report breaks the result out per dimension across the 14-point model.