Phantom Sites and Ground Content for AI Citation Building

AI citations usually resolve to sites most brands have never heard of, but the reason is structural, not accidental. Engines cite what they can corroborate across independent sources, so third-party platforms and community forums win citations by design, while brand-owned pages lose them. If you build ground content and phantom sites deliberately, you address that mechanism directly, but only when the corroboration you manufacture is true.

The overwhelming share of AI citations go to earned, third-party sources: review platforms, community forums, directories, and editorial coverage rather than brand-owned pages. That pattern is not the result of engines preferring popular destinations. This is how retrieval-augmented generation works, and it explains why obscure sites routinely out-cite household names.

AI engines cite what they can corroborate across independent sources, not what is most authoritative in a traditional SEO sense. The underlying mechanism is RAG: retrieval-augmented generation cross-references claims across multiple documents before committing to naming something. Corroboration is the engine's hedge against hallucination, and a single domain, no matter how credible, cannot supply it.

This is why a brand can be named in AI answer text without any of its pages being cited as a source, and why a content aggregator can supply a citation without ever being named. Mentions and citations are different signals flowing through different channels. The same research introduced the term "ghost citation," a URL listed as a source where the brand is never named in the answer, and ghost citations account for the majority of AI citations overall. LoudDemand's 2026 playbook frames the implication directly: brands that are both cited and verifiable win. Showing up in AI answers is only half the job. The other half is showing up in the places people use to fact-check what the AI told them.

The contraction of the crawlable web sharpens the pattern further. Few people recognize them, but the remaining crawlable third-party sources become proportionally more influential as a result. A brand that only publishes on its own domain is structurally underrepresented, regardless of how good its content is.

Phantom sites and seeded ground content function by manufacturing the distributed, independent-source corroboration that AI engines require before they will name a brand. Seeded ground content is accurate brand information placed on legitimate third-party platforms: review sites, directories, community forums, and editorial outlets, all of which expand the citation surface area a brand occupies. Phantom sites are purpose-built domains created primarily to supply the appearance of independent corroboration, but they can be legitimate topic-authority micro-sites or entirely fabricated brand presences.

The experiments confirm the mechanic works, and they also reveal where it breaks. In March 2025, Reboot Online ran an experiment after observing that the agency was not mentioned in ChatGPT's response to the prompt "what are the top 10 best digital PR agencies." ChatGPT recommended it in browsing-enabled answers within three weeks, but Claude and Gemini did not mention it.

The Reboot and Morrowen experiments are different in kind, not just degree. Morrowen created a non-existent product. Both worked, and that gap is the crux of the legitimacy argument that follows.

Whether phantom sites and seeded ground content are a legitimate citation strategy

Legitimacy in ground-content strategy is not a yes-or-no question. It is a function of whether the corroboration being built is accurate, independently verifiable, and durable. Fabricated corroboration exploits the same mechanism as legitimate strategy, so the distinction matters, but it is not a reason to dismiss the practice wholesale.

Practitioners by and large hold that seeding accurate content about a real brand across review platforms, community forums, industry directories, and editorial coverage is modern digital PR adapted for AI retrieval surfaces. That view has genuine merit. Frase's GEO guide notes that GEO overlaps heavily with PR, because most citations go to earned media. The strategy works because it aligns with how the engines already evaluate claims.

The strongest objection is not abstract authenticity. It is structural: AI engines reward entity coherence over truth verification, which means fabricated or misleading content can game the system just as effectively as accurate content, at least in the short term. That is the problem, and it deepens at scale. If brands manufacture "independent" corroboration, the corroboration-as-verification signal collapses. The engine's hallucination hedge becomes the attack surface, and the practitioner defense that this is just earned media provides no guardrail against the Morrowen scenario.

The honest tier distinction resolves the question. Tier 1 is legitimate ground content: accurate brand information distributed across review platforms, community forums, industry directories, and editorial coverage. Tier 2 is structured phantom micro-sites: purpose-built topic-authority sites that publish accurate, independently useful content that happens to mention and describe the brand. Tier 3 is fabricated corroboration: sites or content built primarily to manufacture the appearance of independent verification for claims that either don't exist or couldn't earn genuine coverage. The experiments confirm this works short-term, but durability is unconfirmed and the systemic risk is real.

A governance dimension is emerging alongside these tiers. That raises the stakes for how they are used. News publishers expect search traffic to fall another 43 percent on average over the next three years, and that will only accelerate the shift toward whatever sources remain crawlable.

What this means for how brands should build their AI-visible presence

A durable AI citation presence is built in three coordinated layers: on-site content architecture, distributed ground content on legitimate third-party platforms, and structured phantom or micro-site infrastructure. All of it requires measurement to know whether any of it is working. The first layer is on-site content architecture optimized for extraction. Frase's GEO guide draws the key SEO-to-GEO distinction: SEO rewards long-form pages that cover a topic, while GEO rewards self-contained paragraphs that can be extracted and attributed. Pages that rank for those sub-queries are substantially more likely to be cited, which argues for topically specific pages rather than single comprehensive ones.

The second layer is distributed ground content on legitimate third-party platforms. Review platforms such as G2, Capterra, and Trustpilot, community forums like Reddit, LinkedIn, and Quora, and niche directories represent the source types different engines favor. ChatGPT leans toward editorial and reference sources such as Wikipedia, Reuters, and Forbes. If different engines draw from different source types, you should match platform coverage to the engines your brand most needs to appear in. For a brand's PR and content distribution strategy, the practical implication is to plan around AI citation surface area, not just link acquisition or share of voice in traditional media.

The third layer is phantom or micro-site infrastructure, and it has a specific place within the tier framework. Purpose-built sites that publish accurate, independently useful content on topics adjacent to a brand's domain can legitimately expand the citation corpus, if the content would earn its place without the brand angle. The Reboot Online experiment is the right model: seeded accurate content about a real organization on a domain it controlled, with content that was genuinely answerable and list-formatted in a way that served the extraction behavior AI engines favor. The Morrowen model, a site existing solely to manufacture the appearance of a brand that doesn't exist, is the failure mode. It works until it doesn't, and it degrades the verification signal for everyone.

Measurement closes the loop, and most brands are running blind here. GoodFirms' 2026 survey found that only a minority of marketers track AI or LLM citation visibility, even though a large share name AI optimization as a core strategy. The measurement layer must run continuously to catch drift and trigger republication or distribution on the right platforms. Google AI Overviews appear on 16 to 25 percent of US queries, which makes the surface area too large to leave unmonitored.

Sound methodology matters as much as the tracking itself. Blending creates artifacts that mask what is happening on each platform. Platforms like Letterstory, which monitor AI citation behavior across ChatGPT, Claude, Gemini, and Perplexity, can track whether seeded content and phantom sites are actually moving the needle by measuring not just whether a brand appears in an AI answer but whether it is being named and cited as a source. That is the real signal that ground content is working as corroboration infrastructure.