Generative Engine Ranking Factors
GE ranking factors aren't backlinks and titles. They're fan-out matching, citation density, freshness, and entity match. Different signals entirely. The full treatment lives in generative engine optimization guide.
the Google AI Overviews guide through the Bing Copilot guide covered the major AI surfaces individually — Google AIO, Google AI Mode, Microsoft Bing + Copilot. This page is the cross-engine synthesis. On May 7, 2026, Cyrus Shepard (founder of Zyppy SEO) published a scored analysis of 23 AI citation ranking factors on his Zyppy Signal Substack newsletter, distilling findings from 54 experiments, patents, and case studies published over the past two years across ChatGPT, Gemini, and Perplexity. The top five factors by score: URL accessibility, search rank, fan-out rank, preview controls, and query alignment. The bottom of the table: LLMs.txt — the strongest "what NOT to invest in" signal in the corpus. But scores aren't uniform across engines. Large-scale citation analyses show a heavily concentrated citation economy: a small set of domains captures a disproportionate share of all AI citations. Cross-engine brand recommendations still disagree more often than they agree — each engine builds its own picture of a category. Inside this guide: the 23-factor scored hierarchy with tier bands, per-engine weight differences across six surfaces (ChatGPT / Perplexity / Gemini / AIO / AI Mode / Claude), the fan-out rank case (a six-figure-URL independent study showing a major citation lift), the revenue connection (independent analysis tying AI citations to a substantial organic-click lift per impression), six antipatterns, and the 90-day cross-engine optimization workflow.
The 23-factor scored citation framework
Per PPC.land's published May 7, 2026 coverage of Cyrus Shepard's Zyppy Signal analysis: ("The top factors are URL accessibility, search rank, fan-out rank, preview controls, and query alignment. The lowest-scoring factor is LLMs.txt. The overall finding is that traditional SEO fundamentals remain...") The 12 researchers acknowledged by Shepard include Dan Petrovic, Rand Fishkin, Kevin Indig, Lily Ray, Britney Muller, Ann Smarty, Andrea Volpini, David McSweeney, Metehan Yesilyurt, Michael King, Dawn Anderson, and Oshen Davidson. The factor bars below visualize the top-and-bottom of the 23-factor hierarchy in published score order — banded by tier (TOP / STRONG / MID / WEAK / BOTTOM). The deeper mechanics sit in brand citation tracking.
// Cyrus Shepard 23-factor citation ranking
The 5.4 "known source" finding is operationally important. Per PPC.land's coverage of Shepard: ("Known source scores 5.4 — the phenomenon where an AI system cites a URL from its own training data, bypassing the usual grounding process. This can lead to citations that no longer exist, and it's more typical of ChatGPT and Perplexity than Gemini.") The known-source pathway means content published before training cutoffs accumulates citation probability independent of current rankings — which is why brands with 5+ year content libraries continue earning citations even when their current ranking has slipped. The implication for partner programs: publishing authoritative content early is a compounding investment that survives ranking volatility.
Per Search Engine Journal's published May 14, 2026 methodology commentary on Shepard's study: ("The study is excellent, and the correlation/causation caveat is one he makes himself explicitly... Not only the distinction between what is a factor, and what is a correlation, especially given a lot of studies in SEO and AI are multifarious and have high levels of imponderable complexity.") The 23 factors are correlation, not causation — but when 54 independent experiments converge on the same signals, the operational direction is reliable even when individual causality remains uncertain. The 13.7% overlap measurement makes this a per-surface decision, and the factor weights diverge accordingly.
Six engines, six different optimization profiles
Per Megrisoft's published 2026 cross-engine analysis: ("ChatGPT, Perplexity, Gemini, Google AI Overviews, and Claude all cite differently. Their training data differs. Their grounding behavior differs. Their preferences for source types differ. Treating 'AI search' as a single channel obscures what's working and what isn't.")Large-scale citation analyses show a heavily concentrated citation economy: a small set of domains captures a disproportionate share of all AI citations. For the wider frame, start with ChatGPT vs perplexity vs gemini comparison.
ChatGPT Search — hybrid model with selective retrieval
Cites 3-5 sources per answer. Hybrid of training data (known-source pathway) and selective web retrieval (~31% of prompts trigger web search per Nectiv). Wikipedia 7.8% top source. Reddit 1.8%. Top citations include Wikipedia, Reddit, Forbes, G2. ~90% of ChatGPT Search citations come from pages ranking position 21+ per Semrush July 2025 — substantially lower than Google ranking floor. (Semrush AI search traffic study) Ahrefs' September 2025 dataset — 540,000 query pairs — is the reference measurement behind this behavior. The discipline is unglamorous and the compounding is not — the usual shape of durable advantages. The pattern rewards teams that write the measurement plan before the content plan, never the reverse.
→ Optimize: brand-managed sources + Wikipedia presence + early publication for training data inclusionPerplexity — real-time retrieval with high source density
Performs a real-time web search for every query. Cites 5-10 sources per answer (~3× more than ChatGPT). 91% domain overlap with Google's top-10 per Semrush AI Mode study — most aligned with traditional rankings. Heavy news + academic source preference. High freshness sensitivity.Large-scale citation analyses show a heavily concentrated citation economy: a small set of domains captures a disproportionate share of all AI citations. At 45.5% citation churn per answer update, maintenance cadence outranks launch quality as the durable factor. Instrument it first, then build; movement you cannot attribute is movement you cannot defend. Engineers reviewing the output should recognize their own vocabulary in it; when they do, the engines do too.
→ Optimize: traditional top-10 ranking + freshness + news/academic citations + clean passage extractionGemini — Google's top-10 index dependency
Draws heavily from Google's existing search index. Pages ranking in Google top-10 are far more likely to be cited. In operator practice: visual evidence + structured data alignment increases citation confidence. Million-token context window but still extracts citations from concise self-contained passages per Verge 2025. UGC plays a minimal role in its citation mix — a fraction of what conversational surfaces like AI Mode draw on. (Google Search Central) The 86% semantic-agreement figure means the answer converges while the credit does not — placement is the contest. Teams that skip this step relearn it from their citation-share charts within two quarters.
→ Optimize: Google top-10 ranking + structured data + 134-167 word passages per the Google AI Overviews guideGoogle AI Overviews — 92.36% top-10 dependency
Always-on snippet panel above SERP.The overwhelming majority of AI Overview citations come from domains already ranking in Google’s top ten — AIO visibility is earned through organic authority first. Semantic completeness r=0.87 is #1 ranking factor. 76% to 38% top-10 citation share has compressed over 16 months as Gemini 3 expanded the source pool. Coverage approaches 48% of US queries April 2026. Google's February 2026 documentation state is the compliance reference every structural factor maps back to. The check costs minutes before deploy and the omission costs eligibility after it. Run the check before deploy and after — the state that matters is the one the crawler sees.
→ Optimize per the Google AI Overviews guide — 134-167 word passages + 8.5+ semantic completeness + top-10 floorGoogle AI Mode — lowest top-10 dependence of the Google surfaces
Opt-in conversational tab. Opt-in conversational surface with the loosest tie to classic rankings: AI Mode sidebar sources overlap Google's organic top-10 at just 51% of domains and 32% of URLs — far below AI Overviews. (Semrush July 2025 AI Mode study) It also cites almost every answer (only 3% of responses carry no citations, vs 11% for AIO) and leans harder on Wikipedia (28.9% of citations vs 18.1%). (Ahrefs, December 2025) Multi-turn journey requires 6-module ladder per Section 05. Read against Adobe's 42% conversion advantage for AI-referred visitors, the factor work prices itself. That asymmetry — cheap to maintain, expensive to recover — is the entire argument for the cadence.
→ Optimize per the Google AI Mode guide — module ladder + Reddit/Quora presence + freshnessClaude — training-data heavy with selective retrieval
In observed cross-engine behavior: Claude leans harder on established publishers and training-data authority than its peers, and gives topical-authority content a smaller citation share than ChatGPT or Perplexity. Real-time retrieval enabled but not the default behavior. Smaller training corpus than ChatGPT/Gemini. A materially smaller monthly-user base than ChatGPT or Gemini.
→ Optimize: established publisher citations + brand mentions in news + training-data-era authorityThe operational consequence: cross-surface citation overlap is real but materially incomplete — Semrush's published AI Mode research quantifies how much visibility each surface pair actually shares. (Semrush AI Mode comparison study)
| Perplexity ↔ Google top-10 (URL overlap) | 82% | Semrush queries | Semrush · July 2025 AI Mode study |
| Google AIO ↔ Google top-10 (domain overlap) | 86% | Semrush queries | Semrush · July 2025 AI Mode study |
| Google AIO ↔ Google top-10 (URL overlap) | 67% | Semrush queries | Semrush · July 2025 AI Mode study |
| Google AI Mode sidebar ↔ Google top-10 (URL) | 32% | Semrush queries | Semrush · July 2025 AI Mode study |
| AIO citations from Google top-20 organic | 97% | 432,000 keywords | SeoClarity · 2025-2026 |
| Brand disagreement across ChatGPT / AI Mode / AIO | 62% | Semrush AI & SEO Report | Semrush AI & SEO Report 2026 |
The asymmetry is structurally important. Perplexity has 91% domain overlap with Google top-10 — so a brand ranking well in Google has high probability of being cited by Perplexity automatically. But ChatGPT and Perplexity share only 11% of cited domains — being cited by Perplexity does NOT predict being cited by ChatGPT. (Ahrefs 75K-brand study, Dec 2025) The operational implication: Cross-engine diversification is deliberate, not incidental. answer engine optimization guide carries the end-to-end version. A hostile reviewer tests this factor first; it holds because it is measured mechanism, not a vendor multiplier. What compounds is the system, not the page — partial deployment reads as no deployment.
Fan-out rank: scored 9.3, +161% AIO citation odds
Per Search Engine Land's December 2025 coverage of Surfer SEO's published study: ("An analysis of 10,000 keywords found a strong correlation (Spearman 0.77) between how many fan-out queries a page ranks for and its likelihood of being cited in Google's AI Overviews... Pages ranking for fan-out queries are 161% more likely to be cited than pages ranking only for the main query.") The study extracted 33,000 fan-out queries using Gemini and analyzed 173,902 URLs to derive the correlation. The takeaway: in a world where AI engines decompose user queries into 6-12 sub-queries each, ranking for sub-queries is operationally more valuable than ranking for head queries. The operating detail is in rank in perplexity. The factor stack also inherits from the full service mix — SEO, paid search (Google Ads and SEM), social media marketing, and website design & development — because generative engines read every signal the brand emits, not just the content channel.
The 161% citation lift mechanic
Per SEL: ("Pages ranking for both the main query and at least one fan-out accounted for 51% of AI Overview citations. Pages ranking only for the main query accounted for just under 20%. Fan-outs beat the main query. Ranking for fan-out queries is 49% more likely to earn a citation than ranking for the head term alone.") The asymmetry is structural: AI engines retrieve from sub-queries because that's what the query fan-out mechanic actually retrieves against (per Section 02 live SVG viz). Pages optimized for the head query alone have one citation entry point; pages with sub-query coverage have many.
The 27% consistency caveat
Per SEL: ("Ranking for fan-out queries does not guarantee an AI Overview citation. Fan-outs vary by user context and personalization, with only about 27% remaining consistent across repeated runs.") The fan-out queries Google generates for the same user query change ~73% of the time across repeated retrievals — driven by user context (location, device, search history, time of day). The operational implication: don't optimize for a snapshot of fan-out queries; optimize for the topical neighborhood that contains all plausible fan-out variations. Per Surfer SEO: "don't chase fan-out queries. Own the topic. Build deep topical coverage around core subjects, publishing content that naturally answers a wide range of related questions, and letting AI Overviews discover your relevance across different fan-outs."
How fan-out rank operationalizes
Three concrete moves per Allegiant operating discipline. First: for each priority topic, map 8-12 plausible sub-queries via the internal-linking guide hub-and-spoke discipline + Locomotive fan-out tool + manual ChatGPT prompts. Second: ensure dedicated 134-167 word passages per sub-query within the parent page or across sibling cluster pages per Section 04 + Section 05 ladder. Third: track ranking position per sub-query in Google Search Console + Bing Webmaster Tools — partner programs that rank top-10 for the head query but outside top-50 for sub-queries are leaving the largest single GEO lever unpulled. Score it against the current site before believing any roadmap, including this one.
Why AI citations correlate with revenue lift
Published cross-engine revenue + CTR data — 2025-2026
The commercial mechanics are measured, not asserted: Adobe's 2026 traffic reporting put AI-referred visitors at a 42% conversion advantage over traditional search ) How that plays in practice is mapped in cited by Claude.
The economic asymmetry is the operational case for cross-engine optimization. Citation gain protects revenue; citation loss compounds revenue loss. Per Ahrefs February 2026: a top-ranked page that loses its AIO citation while AIO continues to appear cedes most of its click-through — the click goes to the citation or nowhere. Pages that retain citation against that headwind hold the only click equity left on the query, and with 45.5% of citations churning on answer updates, retention is an active discipline, not a launch state. The revenue gap between cited and non-cited pages widens as AIO coverage expands toward 60-70% of queries forecast through 2026-2027 per Ahrefs trend analysis.
The conversion-quality signal is directionally consistent across the 150-plus published analyses in the statistics literature — and the only figure this page treats as load-bearing is the one measured at scale: Adobe's 42% conversion advantage for AI-referred visitors. The per-visitor revenue value of AI-referred traffic substantially exceeds equivalent organic traffic — the citation premium isn't only about volume protection, it's about quality lift on the volume that flows through. Wikipedia's 28.9% share of AI Mode citations shows how heavily entity-verifiable corpora weigh in the stack. Treat it as infrastructure, not campaign work; the maintenance cost is small and the decay cost is total.
Six defaults that cap cross-engine eligibility
Investing in LLMs.txt instead of URL accessibility
Google's John Mueller put it flatly in June 2025: "no AI system currently uses llms.txt." (Search Engine Roundtable, June 2025) Meanwhile URL accessibility consistently ranks as the top eligibility factor in practitioner scoring. Partner programs investing editorial time in LLMs.txt maintenance while AI bots can't reach the priority pages have inverted the prioritization. The full treatment lives in cross engine consistency.
Optimizing for head query rank instead of fan-out rank
Per Surfer SEO 173,902-URL study: pages ranking for fan-out queries are 161% more likely to be cited than pages ranking only for the main query. Per Cyrus Shepard: fan-out rank scores 9.3 — the highest-impact controllable cross-engine factor. Programs pursuing head-query rank without sub-query coverage capture only ~20% of AIO citation slots vs 51% for pages with both. (Surfer SEO, Dec 2025)
Treating cross-engine as a Google byproduct
Cross-engine citation analyses consistently find only a small minority of domains cited by both ChatGPT and Perplexity.Cross-engine brand recommendations still disagree more often than they agree — each engine builds its own picture of a category. A brand winning visibility in one AI engine may be completely invisible in another — single-engine optimization misses ~89% of cross-engine citation surface.
Blocking preview controls without measuring citation cost
Preview controls rank among the top handful of factors in practitioner scoring — snippet restrictions directly reduce AI citation probability. Programs deploying nosnippet, data-nosnippet, or max-snippet=0 to protect content from being scraped pay a measurable citation cost across all six engines simultaneously. The protective gain must be weighed against the citation upside it forfeits — AI-referred visitors convert 42% better per Adobe's 2026 reporting, and a blocked snippet exits that stream entirely.
Chasing domain authority instead of per-page signals
In operator practice: Domain Authority's correlation with AIO citation has weakened measurably year over year — practitioner scoring now places it mid-pack, well behind entity and passage factors. Programs pursuing aggregate domain authority while neglecting per-page query-answer match (9.2 score) and content structure (8.5 score) miss the page-level signals that AI engines actually weight.
Ignoring brand mentions / unlinked citations
Per Ahrefs' 75,000-brand analysis: brand mentions in YouTube video titles and transcripts are the single strongest correlating factor with AIO visibility among all signals studied. Per Shepard: brand mentions / unlinked citations score 7.8 — strong tier. Programs measuring AI visibility only through linked citations miss the larger brand-mention pathway through which AI engines build entity recognition.
The 90-day cross-engine baseline-to-lift workflow
Days 1-10 — 23-factor scorecard per priority page
For each priority page, score against the top 10 factors from Section 02 (URL accessibility / search rank / fan-out rank / preview controls / query-answer match / intent-format match / content structure / freshness / brand mentions / YouTube presence). Identify the lowest-scoring 3-5 factors per page — these are the highest-leverage retrofit candidates. The deeper mechanics sit in answer first content architecture.
Days 11-20 — Engine prioritization based on ICP behavior
Per Section 03 6-engine cards: select 2-3 priority engines based on ICP query behavior. B2B programs: ChatGPT (68% B2B researcher weekly usage per LinkedIn 2026) + Perplexity (literature research). DTC: AIO + AI Mode + ChatGPT for product discovery. Local services: AIO + AI Mode + Bing/Copilot. Document the selection rationale and lock the priority engine set.
Days 21-35 — URL accessibility audit + preview control review
Verify all six AI crawlers reach priority pages: Googlebot, Bingbot, GPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot. Audit robots.txt and per-page meta tags for nosnippet / data-nosnippet / max-snippet=0 restrictions. Per Cyrus Shepard top-2 factors (URL accessibility 9.5 + preview controls 9.2): remove restrictions unless documented protective gain. Run server log analysis to confirm successful crawler fetches. (Allegiant methodology)
Days 36-55 — Fan-out rank retrofit per priority topic
Per Section 05 fan-out rank discipline + Surfer SEO 161% lift finding: map 8-12 sub-queries per priority topic via the internal-linking guide hub-and-spoke. Deploy 134-167 word passages per sub-query per Section 04. Track ranking position per sub-query in GSC + BWT. Pages with 8+ sub-query top-50 ranking accumulate 51% of AIO citation slots vs ~20% for head-query-only pages.
Days 56-70 — Brand mentions + YouTube transcript discipline
Per the Section 03 tiering, this window builds the Tier 3 surfaces — YouTube, Reddit, and Quora — where brand mentions and transcript-sourced citations accumulate slowest but persist longest, which is why they start mid-program rather than at the end.
Days 71-90 — Cross-engine citation rate measurement
Compare citation rates against day 1-10 baseline across the 2-3 priority engines. Per Section 04 overlap matrix: expect cross-engine variance, not uniform lift. AIO + AI Mode citations reflect at 4-8 weeks per the 60-word answer block pattern platform timing; Bing/Copilot reflects at 1-7 days via IndexNow per Section 06; ChatGPT real-time retrieval ~31% of prompts per Nectiv. Lock weekly engine-specific tracker cadence + monthly 23-factor re-audit + quarterly priority engine reselection.
The 90-day workflow above is the cross-engine synthesis of the workflows in the Google AIO guide, the Google AI Mode guide, and the Bing & Copilot guide. Partner programs running all four workflows in parallel share infrastructure (URL accessibility, fan-out rank, preview controls, brand mentions, YouTube transcripts) but diverge in execution (engine-specific module ladders, IndexNow vs GSC, BWT vs query trackers). Allegiant's discipline: schedule the four workflows as overlapping workstreams with shared infrastructure phases and engine-specific execution phases — the 23-factor scorecard from this page is the unified scoring layer underneath all four.
Written by Chad Markham, President and CEO of Allegiant Digital Marketing. Chad has more than 25 years in digital marketing, including 17 years at a national agency and five years as an instructor in the Digital Marketing program at the University of Texas at Austin. Allegiant is a Google Partner, a Semrush Certified Agency, CallRail Certified, an Inc. Power Partner for 2025, and a 50PROS Top 10 Global agency, serving partners across the United States and Canada. For the wider frame, start with about Allegiant.
Research underneath cross-engine ranking
Every benchmark on this page traces to an independent published source. References below — 8 publications spanning Cyrus Shepard's foundational May 7, 2026 23-factor study, Surfer SEO's 173,902-URL fan-out research, and large-scale cross-engine citation analyses: multi source citation network carries the end-to-end version. (Surfer SEO, Dec 2025)
Get a cross-engine ranking audit on your top 25
Request your free A.R.C. Report. We'll score your priority pages against Cyrus Shepard's 23-factor framework, audit URL accessibility across the 9 major AI crawlers (Googlebot, Bingbot, GPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot, Google-Extended, Bravebot, Perplexity-User), map 8-12 sub-queries per priority topic for the fan-out rank discipline, audit preview-control tag deployment, document existing YouTube + Reddit + Quora brand mention coverage, and deliver the Section 08 90-day cross-engine workflow customized to your current state. Delivered as a custom branded report in 7 business days. No engagement required.
Questions about Generative Engine Ranking Factors
Q What is the 23-factor scored citation framework?
Per PPC.land's published May 7, 2026 coverage of Cyrus Shepard's Zyppy Signal analysis: ("The top factors are URL accessibility, search rank, fan-out rank, preview controls, and query alignment. The lowest-scoring factor is LLMs.txt. The overall finding is that traditional SEO fundamentals remain... The framework's working value is triage: score the twenty-three factors against the current site, rank by gap-times-weight, and sequence the quarter's work by that ranking rather than by instinct — the scoring session takes an afternoon and replaces a quarter of guessing.
Q What are the six engines, six different optimization profiles?
Per Megrisoft's published 2026 cross-engine analysis: ("ChatGPT, Perplexity, Gemini, Google AI Overviews, and Claude all cite differently. Their training data differs. Their grounding behavior differs. Their preferences for source types diffe. Treating 'AI search' as a single channel obscures what's working and what isn't.") The 6-engine optimization profile cards below document the per-engine weighting patterns observed in large-scale multi-engine response analyses, cross-engine benchmark reporting, and large-scale cross-engine citation analyses. The tier logic mirrors the measured selection floor: the overwhelming majority of AIO citations come from top-ten organic domains, so organic authority is the entry fee and the AI-specific factors decide placement among the eligible.
Q What is cross-engine citation diversification is separate work?
Per large-scale published cross-engine analyses: cross-engine citation overlap is structurally low — most domains cited by one engine are not cited by the others. On the Google side specifically, Semrush measured 88% domain overlap but only 58% URL overlap between AI Overviews and AI Mode citations — similar conclusions, different sources. Structural discipline is the controllable half: passage-level answers in the sub-query's vocabulary, extractable first sentences, and sources inside the passage — the shape Google's helpful-content documentation rewards and the retrieval layer selects.
Q Which of the 23 factors should a small business tackle first?
The top-scored fundamentals it already half-owns: crawlability, entity clarity, extractable answers, and corroborating mentions — in that order, because each unlocks the next. The people-first fundamentals score highest across the framework, and the hub's pillar guides document the execution for each.The overall finding is that traditional SEO fundamentals remain... The freshness factor compounds with the churn data: at 45.5% citation turnover per answer update, a stale page is not holding position — it is exiting on a measurable schedule.
Q Is LLMs.txt worth deploying despite its low score?
At 2.0 it is the framework's weakest factor — evidence of adoption by engines is thin. But it costs minutes, carries no risk, and positions the site for whichever engines adopt it. Deploy it as a cheap option, not a strategy; the documented fundamentals are where the actual scores live.The overall finding is that traditional SEO fundamentals remain... The commercial anchor for the whole stack: Adobe measured AI-referred visitors converting 42% better than traditional search — the factors are not academic; they gate a higher-quality visitor stream.
Q How stable are these factor scores?
The fundamentals tier has held stable across every published study cycle; the platform-specific tiers move as engines change grounding behavior. The aggregated dat is re-verified on the page's review cycle, and the weekly measurement is the early-warning system for when a factor's weight shifts in the wild.The overall finding is that traditional SEO fundamentals remain... Cross-engine divergence is the reason single-engine optimization fails: 13.7% URL overlap between Google's own two surfaces means each engine's factor weights get their own treatment — tracked per engine in the citation tracking discipline.
Q How stable are generative engine ranking factors over time?
Unstable by design — and that instability is the planning constraint. Ahrefs measured 45.5% of AI Overviews citations changing when answers update, and the factor weights themselves shift with each engine release. The operational answer is a standing cadence: a fixed prompt panel measured weekly, content refreshed against the factors that moved, and schema re-validated on every deploy. Programs built as launch-and-hold decay on the same curve the churn data predicts; programs built as cadence compound instead.
Q Which ranking factor should a growth-stage firm prioritize first?
Entity grounding, because every other factor inherits from it: an engine that cannot verify who you are discounts everything else you publish. With 77% of B2B buying processes now involving AI per Dentsu's 6,107-interview index, the verification layer is where the buyer meets you first. Baseline current per-engine citation share in an A.R.C. Report, fix the entity graph, then work down the factor stack in coverage order, re-scoring quarterly so the sequence tracks the engines rather than the original plan.