Generative Engine Ranking Factors 2026: 23 Factors Field-Tested Across 5 Client GEO Programs

The GEO conversation in 2026 looks like the early SEO conversation in 2005. Confident opinions, thin evidence, a lot of conference talks. The two pieces of work that cut through the noise are Cyrus Shepard's 23-factor synthesis (Zyppy, May 2026) and the field data accumulating inside agency client programs. This report bridges them.

We took Cyrus's 23-factor framework, scored each factor against our own client outcomes across 5 GEO programs running for 12 months, and added 5 factors we observed in the field that the synthesis missed. The result is the most actionable AI citation factor list we can build with what we know in May 2026.

The dataset is published under a Creative Commons Attribution 4.0 International license. The citation block at the bottom has APA, BibTeX, and HTML formats ready to paste into a draft.

Key findings

Seven stats that summarize the report. Each one stands on its own.

Methodology

The framework is Cyrus Shepard's. The field validation is ours. The two layers are separable. Anyone who wants to cite only Cyrus's framework can do so directly at the Zyppy Signal source. Anyone who wants to cite ProCloser's field test can do so at this URL.

• Framework source: Cyrus Shepard (Zyppy), May 7, 2026. Synthesized 54 experiments, studies, and patents on AI citation behavior across ChatGPT, Gemini, and Perplexity. Each factor scored 0 to 10 on three criteria: repeatability across studies, strength of evidence, official documentation or patent support.
• Validation source: 5 ProCloser.ai client GEO engagements running from May 2025 through April 2026 across B2B SaaS (sports tech, home services), FinTech / B2B financial services, DTC eCommerce, and Financial Services / M&A advisory verticals.
• Measurement stack: TrustRank citation tracking against a fixed monthly prompt set per client across ChatGPT, Perplexity, Google AI Mode, Microsoft Copilot, and Gemini. Per-factor implementation timestamps logged to enable before-after attribution. GA4 referral attribution layered on top to confirm citation-to-traffic conversion.
• Per-factor scoring: Each of the 23 factors was implemented or audited as a standalone change in at least 2 of the 5 client programs over the 12-month window. ProCloser observed effect size scored ordinally as High / Medium / Low / None. Directional consistency scored as the fraction of clients showing lift in the expected direction (e.g., 4 of 5).
• Important caveat: The 5-client sample is too small for statistical significance testing. The ProCloser scores represent field-test directionality, not hypothesis testing. They are intended to flag where the original Cyrus framework either holds, breaks, or needs an extension based on real client work.

Methodology questions can be sent to ProCloser.ai contact. The full per-client implementation log is available on request for journalists, researchers, and partners.

1. The 23 factors, scored two ways

Each factor below shows Cyrus's original score from the Zyppy synthesis and ProCloser's field-observed effect size. The table is ordered by Cyrus score.

Three takeaways from the side-by-side. First, the original framework holds. The factors Cyrus scored 8.5 and above are the same factors we saw consistently move client citations. Second, we scored two factors materially higher than Cyrus: Freshness (we observed bigger effect when date-stamped, see new factor #3 below) and Entity Consistency (we saw direct attribution failures fix when consistent naming was enforced). Third, we scored two factors lower: Length and Domain Authority. Length only matters insofar as it correlates with depth and structure; padding does not help. Domain Authority is heavily proxied by Search Rank, which already does the work.

2. Five additional ranking factors observed in field work

Five factors emerged consistently in our client work but were not in the original 23. We score each ordinally on field-observed effect size.

i. Third-party citation density (effect size: High)

The single largest factor moving generic-query inclusion was the count of third-party citations (review aggregators, listicle blogs, YouTube listicle videos) naming the brand. Owned content cannot fill this gap. Across the 5 clients, the two clients who entered the GEO program with deep aggregator and listicle citation infrastructure already in place reached AI Mode inclusion 60 to 90 days faster than the three clients who started with thin third-party citations. See our companion report on the AI Mode citation source mix for the underlying data.

ii. YouTube listicle inclusion (effect size: High for generic queries, Medium overall)

YouTube tied Reddit for #4 cited source in our AI Mode dataset at 8 citations each. Three of our clients earned AI Mode mentions purely from being featured in a third-party YouTube listicle video ("Top 5 [Category] Agencies 2026") with no other discernable citation source on that query. Producing a single well-shot "Top 10 [Category] [Vertical]" video on the brand's own channel, with honest evaluation criteria placing the brand at the top, moved generic-query citation share in 4 to 8 weeks.

iii. Date-stamped freshness (effect size: High)

The Cyrus "Freshness" factor scored 7.0 based on currency of information. Our field data suggests a sharper version of the same factor. Pages with an explicit month-and-year stamp ("Last updated: May 2026") visible in the first 50 words got cited at materially higher rates than pages with only a year stamp or a "recently updated" tag. The implementation took 30 minutes per page. The lift appeared within 2 weeks on live-web models. We did not find this distinction in Cyrus's underlying studies because the studies treated freshness as a binary or ordinal signal rather than measuring date-format granularity.

iv. Entity-naming consistency across third-party properties (effect size: High)

Brands with consistent naming across all third-party properties (Clutch listing matches G2 listing matches Crunchbase matches Wikipedia matches LinkedIn) got attributed accurately by AI Mode. Brands with name drift (Inc. vs LLC vs "& Co" vs the parent brand) had citations split across entity variants, with the model occasionally attributing citations to the wrong entity or refusing to merge them. Cyrus's framework scored Entity Consistency 5.8. Our field test scored it materially higher because the failure mode is severe: citations exist but do not aggregate.

v. On-page citation infrastructure (effect size: Medium, indirect)

Pages with a visible "How to cite this study" block (APA, BibTeX, HTML blockquote, direct quote) got cited at higher rates by external publishers, which in turn fed third-party citation density (factor i above). This is an indirect effect on AI citation, not a direct one. The page itself does not cite better; the page makes external citation easier, which compounds into more third-party density, which lifts AI citations downstream. Worth budgeting on any page intended as a linkable asset.

3. The four overarching themes

Cyrus's framework groups the 23 factors into four themes. The themes hold against our field data with one ProCloser addition.

The fifth theme (Distribution) is the gap most agencies still have. An on-page audit can score perfectly on Relevance, Trust, Topical Authority, and Extractability, and still fail to move generic-query AI citation rate because the brand is not in the third-party citation network. Distribution is the bridge.

4. Three myths the data does not support

Myth 1: LLMs.txt is the new robots.txt

The narrative is that publishing an llms.txt file at the site root gives AI engines an authoritative content map and lifts citation rates. Cyrus's synthesis found no credible evidence in 54 studies and scored llms.txt 2.0. Our field test agreed. We published llms.txt at 4 of 5 client sites in early 2026 and observed no measurable lift on citation rate attributable to the file. We are not arguing against publishing it. The cost is zero. We are arguing against budgeting implementation effort against it when the same effort spent on third-party citation density or explicit phrasing would move outcomes.

Myth 2: Schema markup is the AI search unlock

Cyrus scored Structured Data 5.6 (modest positive). Our field test scored it Low. Schema is foundational hygiene and a Google requirement for rich results, but it is not the AI citation lever it is sometimes pitched as. Sites with comprehensive schema and weak third-party citation density did not outperform sites with minimal schema and strong third-party citation density on AI Mode inclusion in our dataset.

Myth 3: AI search has replaced classic SEO

The opposite. Classic SEO is the foundation. Top-10 Google rank is the single highest-impact AI citation factor. Ahrefs found that 38% of AI Overviews citations come from top-10 Google results. Sites with weak organic positions cannot compensate by stacking GEO tactics. The right framing is that AI search rewards good SEO and punishes neglected SEO. The two are complementary, not competing.

5. The five-step playbook this data supports

If you run a GEO program based on these 28 factors (23 original plus 5 new), the prioritization the field data supports is the following sequence.

1. Audit and fix Extractability factors first. Strip hedge language. Add TL;DR boxes. Ensure JS-rendered content is also in HTML. Remove accidental nosnippet directives. Implementation effort: low. Time to lift: 2 to 4 weeks. Expected impact: high.
2. Build out fan-out coverage and topic clusters. Map the supplementary queries an LLM generates for your head terms. Publish or refresh content covering each. Implementation effort: medium to high. Time to lift: 8 to 16 weeks. Expected impact: high.
3. Invest in third-party citation density. Get into Clutch, Semrush agencies, DesignRush, G2, GoodFirms with thick reviews. Pitch the top 3 to 5 vertical listicle blogs in your category. Produce one "Top 10 [Category] Agencies" YouTube video per quarter. Implementation effort: medium to high. Time to lift: 8 to 20 weeks. Expected impact: high.
4. Lock in Entity Consistency. Audit every third-party property for name drift. Standardize on one legal entity, one brand name, one URL. Implementation effort: low. Time to lift: 4 to 8 weeks. Expected impact: medium-high.
5. Add freshness signals and on-page citation infrastructure. Date-stamp every cornerstone page with explicit month and year. Add "How to cite" blocks to data pages. Implementation effort: low. Time to lift: 2 to 8 weeks. Expected impact: medium.

Two things to deprioritize. Do not budget against llms.txt as a citation lever. Do not assume schema alone will move AI citations.

6. Bonus findings worth citing

Eight more standalone-citable stats from the dataset.

1. Across 5 clients, on-page Extractability fixes (explicit phrasing, TL;DR boxes, structure) produced the first measurable citation lift in 2 to 4 weeks.

2. Fan-out coverage build-outs took 8 to 16 weeks to lift citation rate but produced the largest absolute citation gains over the 12-month window.

3. Third-party citation acquisition (aggregators, listicle blogs) took 8 to 20 weeks to lift AI Mode inclusion but was the only lever that moved generic-query citation rates.

4. Cyrus Shepard's synthesis covered 54 experiments, studies, and patents. Full source list at Zyppy Signal.

5. Pages with a visible "Last updated: May 2026" stamp in the first 50 words got cited at materially higher rates than pages with year-only stamps in our field test.

6. 38% of AIO citations come from top-10 Google results per Ahrefs, validating the rank-AI correlation at scale beyond our 5-client sample.

7. AIO citations drive 120% more organic clicks and 41% more paid clicks per Seer Interactive, confirming citation-to-traffic conversion.

8. LLMs.txt published at 4 of 5 client sites in early 2026 produced no measurable citation lift attributable to the file across a 6-month observation window.

How to cite this report

Use any of the formats below. The dataset is published under a Creative Commons Attribution 4.0 International license. Attribution is required, derivatives are allowed. Cite Cyrus Shepard separately for the underlying 23-factor framework.

@techreport{kozar2026gerankingfactors,
  title  = {Generative Engine Ranking Factors 2026: 23 Factors Field-Tested Across 5 Client GEO Programs},
  author = {Kozar, Tania},
  year   = {2026},
  month  = {May},
  institution = {ProCloser.ai},
  url    = {https://procloser.ai/blog/generative-engine-ranking-factors-2026/},
  note   = {Field validation of Cyrus Shepard (Zyppy, May 2026) 23-factor framework across 5 client GEO programs, May 2025 to April 2026}
}

<blockquote>
  Top-10 Google rank, fan-out coverage, and explicit phrasing were the
  highest-impact AI citation ranking factors across a 12-month field test
  of 5 client GEO programs. LLMs.txt produced no measurable effect.
  ProCloser.ai field-validated Cyrus Shepard's 23-factor framework and
  added 5 additional factors observed in client work.
  <cite>Kozar, T. (2026). <a href="https://procloser.ai/blog/generative-engine-ranking-factors-2026/">Generative Engine Ranking Factors 2026</a>. ProCloser.ai.</cite>
</blockquote>

About the data, about ProCloser

ProCloser.ai is a Generative Engine Optimization agency for B2B SaaS, FinTech, eCommerce, and professional services brands. The TrustRank methodology behind this dataset combines prompt-set citation tracking, GA4 referral attribution, and on-site GEO improvements to lift brand visibility inside ChatGPT, Perplexity, Microsoft Copilot, Google AI Mode, and Gemini. Learn more about ProCloser or read about the team at Tania Kozar's profile.

This report builds on Cyrus Shepard's original 23-factor synthesis published at Zyppy Signal on May 7, 2026. Anyone citing the underlying framework should cite Cyrus directly. ProCloser's field validation and 5 additional factors are net-new contributions intended to be cited alongside the original, not as a replacement for it. Methodology questions can be sent to ProCloser.ai contact.

Last updated: May 27, 2026. Author: Tania Kozar, Director of Partnerships at ProCloser.ai. Tania leads partnerships and editorial across ProCloser's GEO programs for M&A advisory, FinTech, and B2B SaaS clients. The 12-month field test behind this report draws on her work managing client engagements alongside the ProCloser analytics team.