A 22,410-domain study argues AI visibility is not an output of your SEO program

Fractl's generative engine optimization brand-visibility study analyses 8,090 keywords across 25 verticals and 22,410 leading domains, with the stated purpose of identifying the signals associated with brand visibility in LLMs and AI Overviews ^[1]. The argument that matters for operators is structural rather than tactical: strong Google rankings remain valuable, but they are not a direct measure of whether a brand is surfaced, cited or recalled in LLM-driven search experiences ^[2].

The mechanism is the reason. Traditional SEO seeks visibility in a results page, while generative systems synthesize answers from the information they retrieve and the sources they treat as useful or authoritative ^[3]. The consequence is that a company can have a sound technical SEO foundation and still need to improve the information signals that support its presence in AI-generated responses ^[4]. A highly ranked page is evidence of search visibility for a query, not proof that the brand will be included in a generated answer ^[5].

Now the caveat, because it constrains what you can do with this. The available research does not support a simple scorecard in which hundreds of real brands are classified as underperforming or overperforming purely on the relationship between Google rankings and LLM recall ^[6]. Scope is broad but shallow per slice: 8,090 keywords spread over 25 verticals averages about 324 keywords per vertical ^[7], which is a reasonable basis for a general claim and a weak basis for a category-level read on your own market.

The adjacent evidence carries similar limits. SE Ranking and Search Engine Land tested a fictional brand through 825 prompts across five AI systems over a month ^[8]; that helps explore how AI systems handle recall, but it is not a broad measurement of real brands against their organic-search performance ^[9]. The GEO 2026 barometer from Reworld MediaConnect and ELMARQ emphasises reputation signals, structured data and presence in authoritative sources as factors in LLM visibility ^[10]. None of this retires the basics. Crawlability, indexability, useful content and clear site architecture remain foundational; technical quality alone may just not capture the full set of signals shaping generative answers ^[11].

What the source proposes instead is an operating model with three connected areas: entity clarity, meaning products, services, executives, locations and core claims described consistently across authoritative owned properties ^[12]; content governance, meaning accurate, attributable information that answers real customer questions without vague promotional language ^[13]; and source and retrieval readiness, meaning an assessment of whether important brand facts appear in credible third-party sources and are structured clearly enough to be retrieved and synthesized ^[14].

That work has a second payoff inside the building. The same discipline applies to enterprise retrieval-augmented generation, where conflicting documentation, outdated policies and unclear ownership degrade retrieval quality even when the organisation holds extensive content ^[15]. Externally, no team controls every model's response; it can only improve the quality and consistency of information on its own properties and in relevant authoritative sources, which requires ownership shared across SEO, content, communications, product, legal and data ^[16].

What to watch: whether Fractl publishes signal-level and per-vertical findings rather than headline scope, since 22,410 domains is a sampling frame and not a result ^[1]; whether anyone repeats the prompt-recall work with real brands instead of an invented one ^[8][9]; and whether reporting stops using rank position as the sole proxy for discoverability in AI search ^[17].