What Actually Gets You Cited in AI Search

The AI citation ranking factors that actually drive visibility in ChatGPT, Gemini, Perplexity, and Google AI Overviews are finally measurable. In May 2026, Cyrus Shepard of Zyppy published a meta-analysis that synthesizes 54 experiments, patents, and case studies into a single scored ranking of 23 factors — the first time the GEO field has weighted its advice by evidence strength rather than opinion.

That matters because most generative engine optimization content lists the same factors without ranking them, leaving practitioners to guess which moves are worth the effort. When you pair the Zyppy scoring with the large-scale correlation studies from Ahrefs — 75,000 brands, then nearly 17 million citations, then 4 million AI Overview URLs — a clear and occasionally surprising hierarchy emerges. Some long-held SEO instincts hold up. Others, including a widely-copied statistic about fresh content, do not.

This guide ranks the signals by evidence, decodes the brand-versus- backlinks correlation that is the most actionable finding in the data, explains the fan-out retrieval shift that rewires citation strategy, and converts it all into an evidence-weighted checklist. Every figure is attributed to its primary study, and one popular number is corrected rather than repeated.

The backbone of this analysis is the AI Citation Ranking Factors study Cyrus Shepard published on Zyppy Signal on May 7, 2026. Rather than asserting which factors matter, it synthesizes 54 experiments, patents, and case studies covering ChatGPT, Gemini, and Perplexity, and scores each of 23 factors on three criteria: repeatability across studies, strength of evidence, and official support from platform documentation or patents. The output is a single 0-to-10 score per factor — a rare attempt to weight GEO advice by how well it is actually supported.

We cross-reference that scoring against four large-scale primary datasets so the conclusions do not rest on one source. Ahrefs contributes three: a 75,000-brand correlation study on what predicts AI Overview visibility, a nearly 17-million-citation study on content freshness, and an 863,000-keyword analysis of where AI Overview citations come from. Seer Interactive contributes a 5.47-million-query study on what citation does to click-through. The combination lets us separate signals that show up in one vendor's blog from signals that replicate across independent datasets.

One framing point before the numbers: nearly every figure below is a study-stated correlation, not proven causation. A high correlation between brand mentions and citations does not prove that adding mentions causes citations; it is equally possible that already-strong brands earn both. The original authors are explicit about this, and we preserve their caveats. Treat the scores and correlations as a prioritized map of where the evidence points, not as guaranteed levers.

Five factors cluster at the top of the Zyppy ranking, and they tell a coherent story: AI engines cite pages they can reach, that already rank, that match the query closely, and whose content they are permitted to preview. None of this is exotic — it is the mechanical reality of how grounding and retrieval work, scored against the weight of the evidence behind each claim.

URL accessibility (9.5) is the floor. As Shepard puts it, a URL typically needs to be accessible and crawlable, either during training or grounding, for an AI engine to cite it. Pages blocked by robots.txt, sitting behind a paywall, or returning errors are effectively invisible. Search rank (9.4) sits just below: the Ahrefs data shows 38% of AI Overview citations come from pages ranking in the top 10, which is why the study distills its whole thesis into a single line — win SEO, win AI citations, most of the time, with extra steps.

Preview control (9.2) is the overlooked one. Pages using a nosnippet meta directive to suppress preview snippets can inadvertently reduce or eliminate their AI citation visibility, because AI engines lean on snippet data during grounding. Several teams that locked down snippets for traditional-SERP reasons are quietly excluding themselves from AI answers. Query-answer match (9.2) rewards pages that directly answer the specific phrasing of a query over content that is merely topically adjacent.

The Zyppy study publishes its scores but not a tidy reference table. Below is the full ranking, grouped into the categories we use when scoping GEO work — Technical, Off-page, Content, and a low-evidence tier — with a one-line action per factor. Scores are Shepard's; the grouping and actions are ours. The pattern to notice: the highest-evidence factors are technical accessibility and off-site brand signals, while several content-formatting tactics that dominate GEO advice score in the merely-good range.

This is the most actionable finding in the entire dataset. Ahrefs analyzed 75,000 brands using Spearman correlation on domains with a Domain Rating above 40 and keywords with at least 800 monthly searches. The result reorders the priority list a decade of SEO built around links: the three signals that correlate most strongly with AI Overview visibility are all off-site brand signals — branded web mentions (0.664), branded anchor texts (0.527), and branded search volume (0.392). Domain Rating lands at 0.326, and number of backlinks comes in last at 0.218.

The disparity is even sharper at the extremes. The top 25% of brands by web mention volume average 169 AI Overview mentions versus just 14 for the next quartile — more than a tenfold gap — while the bottom half by mentions barely register, averaging 0 to 3. This is a winner-takes-most distribution: AI visibility concentrates among brands that have built broad web presence, not merely strong link profiles. A separate analysis by Wellows found Domain Authority's correlation with AI citation likelihood has fallen to roughly r=0.18, down from r=0.43 before 2024, converging on the same conclusion from a different dataset.

The strategic read is a genuine reallocation, not a tweak. If your AI visibility program is still pouring budget into raw link acquisition, the data suggests redirecting toward digital PR, unlinked brand mentions, podcast and publication coverage, and the kind of category presence that grows branded search. Our AI citation visibility audit of 500 SaaS sites found the same pattern in practice — the most-cited brands were rarely the ones with the biggest backlink counts.

Fan-out is the mechanical shift that rewires citation strategy. Instead of running a single query and pulling the top results, AI systems expand a user's question into multiple related sub-queries and retrieve across a wider set of SERP results. A page that ranks for many related queries across a topic cluster therefore gains disproportionate citation exposure — which is exactly why Zyppy scores fan-out rank at 9.3 and topic cluster ranking at 8.9, the highest content-strategy signal in the study.

The Ahrefs citation data makes the consequence concrete. Analyzing 863,000 keywords and 4 million AI Overview URLs, the study found that only 38% of AI Overview citations now come from pages ranking in Google's top 10, down sharply from 76% in mid- 2025. Pages ranking 11 to 100 account for 31.2% of citations, and pages beyond rank 100 account for another 31.0%. In other words, nearly two-thirds of citations now go to pages outside the top 10 — because fan-out reaches deeper into the results for each of its sub-queries.

There is a second-order effect worth naming. YouTube is now the single most-cited domain in AI Overviews, having grown 34% over six months to represent 5.6% of all AI Overview URLs — a direct result of fan-out prioritizing video answers for certain query types. The takeaway for content teams is that breadth across a topic, plus presence on the formats AI engines favor for specific intents, matters more than owning the single top spot for a head term. This is the engine behind our citation pattern study and the reason generative engine optimization (GEO) has shifted from page-level tactics to cluster-level coverage. We dug deeper into the rank shift in our analysis of the 38% citation drop from top-10 pages.

Freshness is where GEO advice most often goes wrong, so it deserves a careful correction. You have likely seen the claim that fresh content gets cited "4.3x more" in AI search. We could not trace that figure to any primary study. It does not appear in the Zyppy analysis, which scores freshness a moderate 7.0/10, and it does not appear in the Ahrefs freshness study either. It surfaces only on secondary aggregation blogs without a published methodology or sample size. Do not repeat the 4.3x number.

The verified figure is more modest and far better supported. Ahrefs measured content age across 16.975 million cited URLs spanning seven platforms — ChatGPT, Perplexity, Gemini, Copilot, and AI Overviews among them. AI-cited content averages 1,064 days old (about 2.9 years) versus 1,432 days (about 3.9 years) for organic Google top-10 results. That is a 25.7% freshness advantage, not a multiplier. Real, replicable, and useful — but a far cry from "4.3x."

The practical guidance survives the correction intact: keeping content current is worth doing, the effect is real, and ChatGPT in particular rewards recency more than Google AI Overviews does. But freshness is a moderate factor (7.0/10), not a dominant one, and it sits well below accessibility, rank, and brand presence in the evidence hierarchy. A content team that prioritizes a constant refresh cadence over fixing crawlability or building brand mentions is optimizing the wrong variable. Correcting the "4.3x" myth is itself a small case study in why evidence-weighted ranking matters more than repeating whatever number went viral.

If accessibility and brand presence determine whether you can be cited, content type determines what actually gets pulled. BuzzStream analyzed 4 million citations from 3,600 AI prompts across 10 industries and found that blog and content pages make up 53.46% of all citations — far ahead of news at 14.09% and social at 8.71%. With brand-owned queries excluded, earned editorial content accounted for roughly 80% of citations. Press releases distributed through syndication, by contrast, accounted for a vanishing 0.04% of the entire dataset.

That last figure is worth sitting with. The press-release-and-wire playbook that still anchors many PR budgets is essentially invisible to AI citation systems, while substantive editorial content — the kind written to answer a question well — does the overwhelming majority of the work. One caveat on the BuzzStream data: it was collected over a single week starting late January 2026, so treat the exact percentages as directional rather than permanent.

There is one more reason to prioritize getting cited beyond the citation itself: it appears to lift clicks. Seer Interactive's April 2026 study of 53 brands across 5.47 million queries found that being cited in AI Overviews is associated with 120% more organic clicks per impression versus not being cited. The authors are careful to flag the causation caveat — stronger brands may simply be more likely to earn both citations and clicks — but the correlation is large enough to make citation a visibility goal worth pursuing, not just a vanity metric.

Most GEO checklists are ungrounded opinion. This one maps each action to the evidence weight behind it, so you spend effort in proportion to what the data supports. Work top to bottom: accessibility and rank are prerequisites, brand and cluster presence are the differentiators, and structure and freshness are refinements — not the place to start.

Looking forward, the trajectory in this data points one direction. As fan-out deepens and AI engines pull from ever further outside the top 10, the single-page, single-keyword model of SEO will keep losing ground to brand-and-cluster strategies. The 38% top-10 overlap that replaced last year's 76% is unlikely to reverse; if anything, expect it to keep falling as retrieval gets more aggressive. Teams that treat AI visibility as a brand-presence and topical-authority problem — not a link-count problem — are positioned for where the evidence is heading, not just where it is today. For the structural side of that work, our answer engine optimization (AEO) fundamentals and brand citation audit checklist translate these signals into a working program, and our agentic SEO service builds the cluster coverage and brand-presence engine the data rewards.