The Language of Modern Search

The condition in which a site holds a strong Google SERP position but receives zero citations across AI-generated answer platforms. The site ranks visibly on traditional search but is functionally absent from the AI retrieval layer that now mediates an increasing majority of user queries. Phantom Ranking is invisible to standard SEO dashboards because it cannot be detected through rankings, traffic, or impressions — the metrics remain stable while AI visibility is zero.

The structural exclusion of a site from AI agent-mediated conversion flows because it has not declared WebMCP tool contracts for its conversion actions. When an AI agent working on behalf of a user attempts to execute a booking, enquiry, or purchase action, it looks for a machine-readable action contract. A site without declared contracts cannot be reached by agent execution regardless of its content quality, authority, or ranking position. Agentic Bypass is not a penalty — it is absence of infrastructure.

A web page or site where schema markup exists in isolation — individual Article, FAQ, or Person blocks that contain no @id cross-referencing to any other entity node. An Entity Orphan cannot be graph-traversed by AI verification systems. Each schema block describes a single element in isolation rather than contributing to a verifiable chain of entity provenance. The result is that AI citation systems treat the page as a lower-trust anonymous source even when the page contains explicit author and organisation markup.

The state in which a site’s core content is rendered client-side through JavaScript frameworks, making it visible to human visitors but invisible to AI inference crawlers that do not execute JavaScript. A site operating in an Inference Shadow may have exceptional content quality, strong backlink profiles, and complete schema markup — all of which are functionally irrelevant because the AI system never received the content. The Inference Shadow is confirmed by the JavaScript-disabled browser test: any content that disappears from view when JavaScript is disabled is inside the shadow.

A specific page or content cluster within an otherwise AI-accessible site that is excluded from retrieval due to a localised technical condition — a page-level noindex tag applied accidentally, a Cloudflare rule scoped to specific URL patterns, a paywall triggered for specific user-agent strings, or interactive tab content that contains the primary argument. Unlike site-wide invisibility, an AI Blindspot affects targeted pages while the rest of the site remains citation-eligible. AI Blindspots are often discovered through citation gap analysis: comparing which pages receive AI citations against which pages hold Google rankings.

The compounding force by which early AI citations produce conditions that make subsequent citations more likely. A source that is cited consistently by one AI platform begins to appear in the training and retrieval data used to calibrate citations on other platforms. Entity reputation signals, unlinked brand mentions, and third-party references accumulate around frequently cited sources. Citation Gravity is directionally similar to domain authority in traditional SEO but operates through a different mechanism: retrieval frequency and cross-platform consistency rather than backlink signals. Early-mover advantage in GEO is largely a function of establishing Citation Gravity before competitive windows close.

The phenomenon in which an AI platform retrieves and uses content from a source without displaying the source’s brand name in the generated response. Research by Superlines (March 2026) documented 182 instances of a site being cited by Gemini in a single month with zero brand name mentions. Ghost Citations create a measurement problem: the site is contributing to AI-generated answers and may be influencing user decisions, but receives no direct attribution and no measurable referral traffic. Ghost Citations are only detectable through systematic manual auditing of AI responses across target queries.

AI-referred website visits that are misattributed as Direct traffic in analytics platforms because the referring AI system does not pass standard referral headers. Research from Wheelhouse DMG (2026) indicated that approximately 91% of actual AI-driven visits appear as unattributed Direct traffic in GA4. Dark Citation Traffic represents a structural measurement failure in which standard analytics dashboards systematically undercount AI search as a discovery channel, causing teams to underinvest in GEO based on data that reflects a fraction of actual AI-driven activity.

The gradual reduction in AI citation frequency experienced by a page as it ages without updates, driven by AI retrieval systems’ preference for recent sources. Research by Amsive (2026) found that 50% of content cited in AI responses is less than 13 weeks old. Unlike traditional SEO ranking decay, which is gradual and driven by competitors accruing backlinks, Retrieval Decay can be rapid and is primarily driven by absolute recency signals rather than relative competitive displacement. A page that earned consistent citations in January may lose most of them by April not because a competitor improved but simply because the content became older than the retrieval system’s recency threshold.

The complete, verifiable sequence of entity cross-references — from Organisation to Person to Service, linked through @id URIs and confirmed through sameAs references to Wikidata or other knowledge graph entries — that enables AI retrieval systems to confirm the identity and authority of a source through graph traversal. A complete Provenance Chain allows an AI system to enter the entity graph from any node (the article, the author, the organisation) and traverse the full relationship. An incomplete Provenance Chain — where individual schema blocks exist without cross-referencing — produces an Entity Orphan condition and reduces citation selection priority.

The process by which readers perform a rapid, pre-conscious assessment of a piece of content to determine whether it matches their psychological arrival state before making a stay-or-leave decision. Psychological Indexing occurs in the first three to five seconds of a page encounter and operates through pattern recognition rather than deliberate content evaluation. Content that fails Psychological Indexing is abandoned before it is read regardless of its technical quality, depth, or accuracy. The five primary arrival states that content must match to pass Psychological Indexing are: Uncertainty Aversion, Loss Aversion, Authority Deference, Cognitive Ease, and Identity Alignment.

The condition in which the register, tone, or opening framing of a page fails to match the dominant psychological state of its arriving visitors. Intent-State Mismatch is distinct from traditional search intent mismatch (where content addresses the wrong query) — it addresses the emotional and cognitive state the user is in when they arrive, not merely the topic they searched. A page can correctly address the searcher’s topic while simultaneously triggering an Intent-State Mismatch through hedging language, abstract openings, or a tone that signals low confidence. Intent-State Mismatch is the most common cause of high bounce rates on technically well-optimised pages.

A page exit triggered not by mismatched topic or intent but by excessive cognitive load — dense prose, complex sentence structures, technical jargon without contextualisation, or visual information density that exceeds the processing capacity the reader is willing to commit. Cognitive Bounce is distinct from standard bounce in that the topic and intent are correctly matched; the reader left because the delivery format made the content too costly to process. Cognitive Bounce is measured by comparing bounce rates against Flesch-Kincaid readability scores across pages covering identical topics. Pages with equivalent topical relevance and meaningfully different readability scores consistently show readability as the primary bounce predictor.

A content architecture approach in which structural and tonal decisions are made based on the documented psychological arrival states of target readers rather than keyword signals or topical coverage requirements alone. Behavioural Indexing treats the human reader as the primary retrieval system that content must satisfy — before and in addition to technical SEO and GEO requirements — on the grounds that content which satisfies human psychological arrival states also produces the engagement signals (dwell time, low bounce, return visits) that reinforce algorithmic and AI citation selection.

A content writing methodology in which the opening 150 to 200 tokens of any page or section are treated as the highest-priority editorial real estate — because AI retrieval summarisation pipelines assign disproportionate weight to this token window when selecting passages for citation. Token-First Architecture requires the first sentence of every major content unit to contain a definitional statement in the form: [Entity] is [category] that [verifiable differentiating claim]. Generic introductory paragraphs, contextual scene-setting, and rhetorical questions in the opening segment are incompatible with Token-First Architecture because they occupy high-weight tokens with zero-information content.

The percentage of manually tested target queries on a given AI platform that return a citation to a specific domain or author within the generated response. CFR is calculated per platform, per query cluster, and per time period to produce a comparable, trackable GEO performance metric. It is distinct from impression data (which is largely unavailable from AI platforms) and from referral traffic (which captures only the subset of citations that result in a click). CFR is currently the most reliable primary metric for GEO performance monitoring.

The gradual divergence between a brand’s actual positioning and the way AI-generated answers represent that brand over time — caused by AI models updating their synthesised understanding of a brand based on third-party sources, forum discussions, and competitor content rather than primary source materials. Answer Drift is undetectable through standard SEO monitoring. It requires periodic manual query testing across AI platforms specifically checking whether the synthesised description of a brand matches the brand’s intended positioning. Uncorrected Answer Drift can result in AI-generated answers that accurately represent a competitor’s framing of a brand.

The percentage of AI-generated responses within a defined topic cluster or query set that include a citation to a specific brand, domain, or author — expressed as a proportion of all citations delivered within that topic cluster across all cited sources. Share of Answer is the GEO equivalent of Share of Voice in traditional media measurement and of organic share of traffic in traditional SEO. A brand with a 35% SoA in the GEO platform layer for its target topic cluster is cited in more than one third of all AI-generated answers in that space across the platforms measured.

The condition in which a source exercises significant influence on AI-generated answers in a topic area through consistent citation but receives no brand attribution, no visible credit in generated responses, and no directly trackable referral traffic. Invisible Authority is the cumulative effect of Ghost Citations and Dark Citation Traffic operating over time. A brand with high Invisible Authority is shaping what AI systems say about its topic area without receiving the brand recognition value of that contribution. Resolving Invisible Authority requires entity schema improvements (strengthening the Provenance Chain), llms.txt deployment to guide inference engines toward brand-attributed pages, and consistent naming conventions across all published content.