Skip to content

Instantly share code, notes, and snippets.

@Coder1400
Created April 13, 2026 04:05
Show Gist options
  • Select an option

  • Save Coder1400/4d9028073779ed28fb5812a3bb15e304 to your computer and use it in GitHub Desktop.

Select an option

Save Coder1400/4d9028073779ed28fb5812a3bb15e304 to your computer and use it in GitHub Desktop.
AEO 3-layer citation classification README (draft for qcomp45/alayah#45)

AEO Citation Classification Model

This document describes how we classify citations returned by AI engines in our AEO product.

Our goal is to turn raw citations (URLs/domains) into structured data that explains:

  • WHO is behind the content
  • WHY / under what incentives the content was created
  • WHAT the page looks like structurally

This 3-layer model lets us analyze citation patterns in a way that existing AEO tools cannot. Most AEO tools today classify citations only at the domain level (for example, Owned vs. Media vs. Social), which loses critical page-level nuance such as affiliate monetization or hybrid editorial content.

We classify every citation independently along three axes:

  1. Source Type — WHO is behind this site?
  2. Content Type (Content Origin) — WHY was this specific page created and under what incentives?
  3. Format Type — WHAT does this page look like structurally?

Layer 1 — Source Type (WHO is behind this site?)

Question: What type of organization is publishing this domain?

Level: Domain-level (applies to all URLs on the same domain)

Definition:
Source Type describes the publisher / domain owner, independent of any particular article or page. It answers: “What kind of entity runs this website?” Examples include: a brand’s own site, a competitor, a news outlet, a review platform, a government agency, or a non-profit.

We use Source Type to understand whose voice AI models are amplifying when they cite content about a brand.

Why this matters

  • Different types of publishers (media, social, institutional, brand-owned, etc.) tend to be cited in different situations by AI engines.
  • Domain type is a low-cost, stable signal we can compute and cache. It is the foundation for understanding the mix of voices in a brand’s citation profile.

Layer 2 — Content Type (Content Origin)

Question: What is the economic and editorial origin of this page?

Level: Page-level (varies across URLs on the same domain)

Definition:
Content Type captures who created this specific piece of content and what incentives or disclosures apply. Examples:

  • Independent editorial coverage written by a journalist.
  • Affiliate content earning commission through outbound links.
  • Brand-authored marketing copy on a product page.
  • User-generated reviews or forum posts.
  • Sponsored posts or press releases.
  • Product documentation or developer docs.
  • Original research reports or benchmark studies.

This is not about the technical act of publishing in a CMS. It is about the business model and editorial origin of the page:

  • Is this editorial coverage written by an independent publisher?
  • Is it affiliate content with commission-based links and disclosures?
  • Is it UGC (user-generated content) on a platform or community?
  • Is it sponsored content or a press release?
  • Is it brand-owned marketing or documentation created by the product team?

We use Content Type to understand the incentive structure and potential bias behind citations.


Layer 3 — Format Type (WHAT does this page look like structurally?)

Question: How is this page structured and presented to the reader?

Level: Page-level

Definition:
Format Type describes the layout and content structure of a specific URL. Examples include:

  • Listicle – "Top 10 Best CRMs" style numbered lists.
  • Comparison – "X vs Y vs Z" side‑by‑side evaluations.
  • Review – single product/service deep‑dive review.
  • Product page – price, features, and conversion CTAs.
  • Landing page – campaign or offer page with one dominant CTA.
  • Category page – collection or listing of multiple products.
  • How‑to / Guide / Educational – structured instructional or explainer content.
  • FAQ / Wiki / Glossary – Q&A, reference, or term‑definition structures.
  • Case study / Research / News / PR – narrative or report‑style formats.
  • Tool – interactive calculators, generators, or checkers.
  • Social thread / Forum thread – conversational posts and replies.

Academic web genre research and real‑world citation data both show that format strongly predicts likelihood of being cited: listicles, comparisons, and guides are disproportionately present in AI answers.

We use Format Type to understand which structural page types AI engines prefer to quote and where a brand may need to produce or improve specific formats.


Summary

The three layers work together as follows:

  • Source Type tells us who is speaking.
  • Content Type tells us why that specific page exists and what incentives shaped it.
  • Format Type tells us how the information is structured for readers and AI engines.

By keeping these layers separate and orthogonal, we can:

  • Avoid forcing a single, lossy label on complex publishers (e.g., Forbes being both editorial and affiliate).
  • Run much richer analyses on which voices, incentive models, and formats dominate a brand’s AI citation profile.
  • Extend the model later (for example, with Brand Framing or trust scores) without breaking the core classification scheme.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment