Source: zyppy-ai-citation-ranking-factors-2026-05-07.md — Cyrus Shepard, Signal by Zyppy (Substack). Published 2026-05-07.

Cyrus Shepard (Zyppy SEO) synthesized 54 published AI-citation experiments, patents, and case studies into a single ranked factor list. The output is a 23-factor table with evidence-based 0-10 scores for ChatGPT, Gemini, and Perplexity. The strongest signal: AI citation engines re-rank on top of classical search relevance, so winning organic SEO is the precondition. The most-hyped 2025 tactics — schema and LLMs.txt — score near the bottom (5.6 and 2.0 respectively). Shepard’s thesis: “win SEO, win AI citations (most of the time, with extra steps).”

Key Takeaways

  • Top tier (9+ across all engines): URL Accessibility (can the bot reach the page?), Search Rank (does it already rank on Google?), Fan-out Rank (does it rank for query-expansions, not just the head term?), Preview Control (does your title/meta-description match what the engine is asked?), Query-Answer Match (does the page literally answer the question?), Intent-Format Match (does the format — listicle, table, definition — match the intent?).
  • Mid tier (7-8): Direct Quote Density, Statistic / Number Specificity, Citation by Other Cited Sources (the “second-order citation” effect), Recency.
  • Low-scoring tactics (under 5): Backlinks (4.8 — present but indirect), Long-Form Content (4.2 — past 1500 words, returns diminish), Brand Mentions Without Link (3.9), LLMs.txt (2.0 — the most overhyped tactic of 2025-26 by Shepard’s read).
  • Structured Data scores 5.6 (#20 of 23). Helps with parseability and rich snippets but doesn’t independently bump citation odds. Empirically corroborated by the Ahrefs schema causal study (2026-05-11) which found no statistically meaningful AI citation lift from adding JSON-LD schema.
  • Per-engine variation matters. Gemini weights Search Rank more than ChatGPT (Gemini is more tightly Google-Search-backed). ChatGPT weights Preview Control and Recency more (RAG over fresh web). Perplexity weights Citation-by-Cited-Sources more (its retrieval is more cluster-based).
  • The single-best-explanation hypothesis. Search Rank, Fan-out Rank, and Query-Answer Match all sit at 9+. These three together describe ~75% of AI citation outcomes per Shepard’s read of the 54 studies. The remaining 25% is engine-specific tweaks (recency for ChatGPT, etc).
  • Why this matters more than individual case studies. Each of the 54 underlying studies has small-sample or selection-bias issues. Aggregating them with weighted scores removes some noise and surfaces the factors that consistently matter across studies, surfaces, and methodologies.

The 23 Ranking Factors (Top-to-Bottom)

#FactorAvg ScoreChatGPTGeminiPerplexity
1URL Accessibility10.0101010
2Search Rank9.791010
3Fan-out Rank9.39109
4Preview Control9.01089
5Query-Answer Match9.0999
6Intent-Format Match9.0999
7Direct Quote Density8.0888
8Statistic / Number Specificity7.7878
9Citation by Other Cited Sources7.3778
10Recency7.0876
11Internal Linking6.7776
12Topical Authority6.7776
13Domain Authority6.3676
14Page Speed6.0666
15Mobile UX6.0666
16Anchor Text5.8665
17Entity Mentions5.7665
18Heading Structure5.7665
19Image Alt Text5.7665
20Structured Data5.6656
21Backlinks4.8554
22Long-Form Content (1500+ words)4.2445
23LLMs.txt2.0222

Scores are averaged across the 54 underlying studies, weighted by study quality (sample size, control design, peer review). Methodology described in full in the source.

Tactical Implications

  • Win SEO first. Search Rank is #2 (9.7 avg). If you aren’t ranking organically, AI citations are a near-impossibility regardless of every other factor. This collapses the “AEO is separate from SEO” narrative.
  • Optimize for query fan-out. Fan-out Rank is #3. AI engines expand the query into sub-queries (the Google AI Mode query fan-out pattern); pages ranking for the head term and its expansions get cited more. Practical move: cluster your content around topical concepts, not just exact-match keywords.
  • Control your preview. Preview Control (#4) means your title and meta description literally appear in the engine’s source-attribution UI. A clear, query-matching preview gets clicked / cited more than a clever-but-vague one.
  • Format-match the intent. A “how to” query wants a numbered listicle. A “what is” query wants a definition + example. A “compare” query wants a table. Wrong format = no citation even if your content is correct.
  • Skip LLMs.txt. Score 2.0 (#23). The most overhyped 2025 tactic. Engines aren’t actually using it as a retrieval signal at meaningful rates.
  • Schema for the right reasons. Score 5.6 (#20). Add schema because Google still rewards it on classical surfaces (rich snippets, knowledge panel) — not because of AI-citation lift. The Ahrefs causal study is the independent empirical confirmation.

Per-Engine Subtleties

  • ChatGPT. Highest weight on Preview Control and Recency. RAG-heavy retrieval over fresh web means your title/preview matters more and stale content gets dropped faster.
  • Gemini. Highest weight on Search Rank. Gemini is the most tightly integrated with Google Search, so winning classical SEO maps almost-directly into Gemini citations.
  • Perplexity. Highest weight on Citation-by-Other-Cited-Sources. Perplexity’s retrieval clusters around already-cited-by-others sources, so the second-order citation effect is strongest here. Implication: getting cited somewhere in the Perplexity-citation graph creates compounding returns.

Open Questions

  • The exact 54 underlying studies. Shepard lists representative examples but the full study list lives in his methodology appendix. Worth pulling for citation-chain verification. ^[inferred]
  • How scores decay over time. AI engine retrieval changes monthly. The 2.0 for LLMs.txt assumes engines aren’t using it as of May 2026 — if any engine starts honoring it, this score would jump.
  • Weighting by traffic / vertical. All 23 factors are averaged across studies. Per-vertical (e.g., medical, legal, e-commerce) weights would likely shift the rankings — the consensus is that medical content gets cited more conservatively, so factors like Topical Authority and Brand Authority probably weight higher there.
  • Ahrefs Schema → AI Citations Causal Study — Independent causal-inference study (1,885 pages DiD) that empirically confirms Shepard’s 5.6 score for Structured Data: no statistically meaningful AI citation lift from adding schema.
  • AirOps + Kevin Indig Fan-Out Effect ChatGPT Study — Largest single-engine empirical dataset. AirOps’s retrieval-rank finding (rank-1 cited 58.4% vs rank-10 14.2%) is the deepest empirical backing for Shepard’s 2-3-5 weights (Search Rank, Fan-out Rank, Query-Answer Match).
  • Digital Applied 1,000 AIO Citation Pattern Study — Concrete AIO-only data on schema lift (2.3×-2.8× after DA control). Magnitude varies between Shepard’s meta-analytic 5.6 and Digital Applied’s much-larger correlational lift — see Open Questions about meta-analysis weighting.
  • GEO-16 Framework (arXiv 2509.10762v1) — Academic cross-sectional study. Likely one of the 54 underlying studies in Shepard’s meta-analysis. ^[inferred]
  • FLUQs Framework — The “core SEO first” thesis. Shepard’s 23 factors are the empirical backing.
  • Google’s Generative AI Search Optimization Guide — Google’s official position aligns: AI Overviews + AI Mode use the same Search index, so AI search is still SEO.
  • Similarweb Most-Cited Domains in LLMs — What domains LLMs actually cite. The factor ranking is how you become a cited domain; Similarweb’s data is who the cited domains are.
  • GSC Autonomous SEO Engine — Operationalizes factors 2-6 (Search Rank, Fan-out Rank, Preview Control, Query-Answer Match, Intent-Format Match) via GSC query data.

Try It

  1. Pull your own top-10 ranking pages from GSC. Cross-reference against the 23 factors. Identify the 1-2 lowest-scoring factors on your pages — those are your highest-leverage improvements.
  2. Audit your title tags and meta descriptions. Preview Control (#4) is one of the cheapest wins. Rewrite any preview that doesn’t literally answer the page’s primary query.
  3. Check fan-out coverage. Run your head term through Google AI Mode and capture the 5-10 expanded sub-queries it generates. Does your page address all of them? If not, expand sections or add an FAQ block.
  4. Stop new LLMs.txt projects. Score 2.0. Reallocate the engineering hours to factors 2-6.
  5. Re-frame schema work. Keep your existing schema (still helps classical Google surfaces). Don’t expand schema in pursuit of AI citations — the Ahrefs causal study + Shepard’s #20 ranking both say it’s not there.