Skip to content
AEO Canon · the reference for answer-engine optimization
Research whitepaper · v1.0 · 2026

The AEO Canon: eight principles for earning AI citation

Abstract

Answer Engine Optimization (AEO) is the discipline of making content discoverable, trustworthy, and citable by AI answer engines. This paper presents the AEO Canon: eight evidence-grounded principles in three layers — Foundation, Reputation, Momentum. The central finding: AI citation is governed by the same broad forces as search quality plus a distinct extraction layer. Tactics widely promoted as essentials — schema markup, llms.txt, keyword optimization — show little to no causal effect in controlled testing. The durable advantages are genuine authority, verifiable credibility, and original contribution.

1 · The search paradigm has shifted

For three decades the defining question was can users find you in search results? The answer was a click. That question has been superseded by another: does the AI cite you when it answers? The user receives an answer, not a menu — the cited source receives the visibility, and every uncited source receives nothing.

Seer Interactive documented organic click-through falling 61% when an AI Overview appears; Ahrefs found cited brands saw 35% more organic and 91% more paid clicks. The penalty for absence is real; the reward for citation compounds. Only 38% of pages cited in AI Overviews still rank in Google’s organic top ten — down from 76% seven months earlier.

2 · Research foundation & methodology

The Canon is built on primary research, weighted by evidence quality. Controlled experiments are trusted over correlation:

  • The Princeton GEO Study (2024) — the only peer-reviewed controlled experiment in the field. Nine content strategies across 10,000 queries.
  • Ahrefs Brand Visibility Studies — 75,000 brands, plus a 1,885-page controlled schema test.
  • SE Ranking — 129,000 domains; Vercel / MERJ — 500M+ GPTBot requests; Semrush — 100M+ citation events.

3 · AEO and SEO: the relationship

AEO does not replace SEO — it extends it. Roughly 70–80% of what drives citation also drives rankings. The genuine divergence is specific: retrieval at the passage level, mentions over links, conversational queries, aggressive freshness weighting, and per-engine divergence (~11% cross-engine overlap).

4 · The Canon framework

Eight principles in three layers. The pillars are sequential: failures at earlier stages make later efforts irrelevant. Walk down it in order; the pillar where you break is the place to work. See the interactive framework.

5 · Layer One — Foundation

1 Access.A page is machine-readable or it is invisible. Vercel/MERJ’s analysis of 500M+ GPTBot requests found zero JavaScript execution — crawlers fetch raw HTML only. 2 Alignment. Search queries were shorthand; AI queries are full conversations — use the question itself as the heading. 3 Extractability. RAG systems retrieve passages, not pages; 44% of ChatGPT citations come from the first third of a page.

6 · Layer Two — Reputation

4 Authority. Determined off-site — branded mentions correlate with AI visibility at 0.664 vs 0.218 for backlinks (Ahrefs, 75k brands). Reddit is the most-cited domain across engines. 5 Credibility. The Princeton GEO study found adding quotations raised visibility 41% — the largest single effect; keyword stuffing performed below baseline. 6 Originality. Original data exists in exactly one place, so the engine cites that place. Your data cannot be scraped.

7 · Layer Three — Momentum

7 Freshness. Seer found 65% of AI crawler visits target content under one year old. 8 Adaptability. Semrush documented ChatGPT’s Reddit citation rate swinging from ~60% to ~10% in six weeks. Measure per engine on a fixed prompt set; the principles endure, the specifics are written in pencil.

8 · What the evidence debunks

Schema markup as a citation driver:Ahrefs’ 1,885-page test found AI Overview citations dropped 4.6% after adding JSON-LD. Valuable for rich results — not a proven citation lever. llms.txt as a citation signal: only 84 of 62,100 AI bot requests (0.1%) fetched it; no engine documents using it. Keyword stuffing: a 10% degradation below baseline.

9 · Measuring AEO performance

There is no Search Console for AI; citation is probabilistic and personalized. The reliable unit is share of voice: a fixed set of 20–50 representative prompts, measured per engine, tracked as a trend, paired with server-log crawl confirmation. Ahrefs found AI-referred visitors converted at ~23× the rate of organic.

10 · Conclusion

The durable way to earn AI citation is to deserve it. Machines designed to find the clearest, most specific, best-evidenced, most-trusted source will, over time, find the sources that are genuinely those things. The eight pillars are not tactics — they are a description of what a genuinely excellent, trustworthy, original source looks like in the age of AI.

The principles endure. The specifics are written in pencil.


aeocanon.com — Version 1.0 · 2026. The AEO Canon is a living framework; principles are updated as new evidence emerges. Sources include the Princeton GEO study (KDD 2024), Ahrefs (75k brands), SE Ranking, Vercel/MERJ, Semrush, Profound, and Seer Interactive.