A living glossary of GEO.

A near-synonym of GEO, emphasising optimisation specifically for direct-answer formats — FAQ boxes, definition cards, AI Overview summaries — rather than citation in conversational AI. Some practitioners treat AEO as a subset of GEO focused on structured, concise answer fragments.

Google's conversational search interface powered by Gemini, launched in 2025. By Q2 2026 it surpassed 1 billion monthly users with queries doubling every quarter. AI Mode replaces the link list with a synthesised answer and source citations. Source: Google I/O 2026.

Google's AI-generated summary that appears above organic results on the SERP, powered by Gemini. As of March 2026, AI Overviews appear on ~48% of commercial-intent queries tracked by BrightEdge, and on 95% of comparison queries (X vs Y format). Brands cited inside an AI Overview earn 35% more organic clicks than non-cited competitors on the same SERP. Source: Digital Applied, March 2026.

A platform that generates synthesised, direct-answer responses to queries rather than returning a ranked list of links. The five major answer engines as of 2026: ChatGPT (OpenAI), Perplexity, Gemini (Google), Copilot (Microsoft), Google AI Overviews. Each uses a different retrieval architecture and citation model.

The binary measure of whether a brand is mentioned at all in an AI-generated response to a given prompt. Answer presence is the baseline metric of AI visibility — a brand is either in the answer or it isn't. Typically expressed as a percentage across a representative sample of target prompts. Different from citation share, which measures relative frequency vs competitors.

An emerging discipline focused on brand visibility when autonomous AI agents browse and transact on behalf of users, without requiring human review of the result. Framed by Adobe in its April 2026 Semrush acquisition as a third layer alongside SEO and GEO. Google launched information agents for Pro/Ultra subscribers at I/O 2026.

An AI response where your brand's domain is explicitly linked or named as the source of specific information — a higher-value signal than a bare mention. On Perplexity, 94% of answers contain inline numbered citations [1][2][3]; first-cited sources capture 48–58% of referral clicks. Source: MarGen, 2026.

The practice of structuring content specifically to maximise citation frequency in AI answers. Includes: sourcing all factual claims with inline hyperlinks (+41% lift), adding specific statistics (+32% lift), including named expert quotations (highest single-signal lift), and FAQ schema markup (+67% LLM discoverability). Source: Princeton/KDD 2024; Yext 2026.

Of all AI citations in a given topic or query cluster, the proportion that reference your brand vs competitors. The GEO equivalent of share of voice. Measured across a representative prompt sample with disclosed methodology. Citation share is directional (trending up or down) and must be tracked against engine-level changes in retrieval behaviour.

The process by which AI retrieval systems extract discrete passages from a longer document to synthesise an answer. Retrieval-Augmented Generation (RAG) systems prefer chunkable content: paragraphs of 3–4 sentences max, clear headings at each H2/H3, tables, FAQ sections. Content that cannot be cleanly chunked is less likely to be cited even if indexed.

A weighted aggregate score (typically 0–100) combining answer presence, citation share, and sentiment quality. The AI Visibility Index methodology weights these 40/35/25 respectively. Composite scores enable cross-domain ranking and trend tracking. A score without disclosed weighting and prompt methodology is not defensible.

The process of ensuring an AI model has sufficient, consistent, structured information about your brand to represent it accurately across all answer contexts. Achieved through: Organization schema markup, Google Knowledge Panel maintenance, consistent entity naming across all web properties, Wikipedia coverage where eligible. Reduces hallucination risk significantly.

Experience, Expertise, Authoritativeness, Trustworthiness — Google's quality rater criteria, which also influence what content AI retrieval systems prefer to cite. AI models trained on web-quality signals absorb E-E-A-T implicitly. Signals include: named authors with verifiable credentials, cited primary sources, institutional affiliations, editorial review processes.

When an AI engine decomposes a single user question into multiple sub-queries, which it resolves independently before synthesising a final answer. Relevant for brands in complex categories: a query like "best CRM for a 10-person agency under $50/seat" becomes 4–8 internal sub-searches. Brands must be present across all sub-query types, not just the aggregate category query.

The recency preference in AI citation selection. 85% of AI citations are from content published or updated within the past 2 years. Recently updated content appears 4.3× more often in AI answers than stale equivalents. Freshness is weighted more heavily by Perplexity (real-time web search) than by ChatGPT (training corpus + optional web). Source: Seer Interactive, 2026.

The practice of optimising content, technical infrastructure and off-site authority signals to improve how often and how accurately AI answer engines retrieve, cite and represent a brand. Term coined in the Princeton/Georgia Tech/IIT Delhi paper "GEO: Generative Engine Optimization" (Aggarwal et al., KDD 2024), which demonstrated citation lifts of up to 41% from specific content interventions. By 2026 it commands a $3.2B market (tooling + services), validated by Adobe's $1.9B acquisition of Semrush (April 2026).

Tying an AI-generated claim to a verifiable, retrievable source document. Grounded answers cite specific domains and pages; ungrounded answers may hallucinate facts. From a brand perspective, grounding works in two directions: making your content the source a model retrieves, and ensuring your brand facts are correctly represented in that retrieval.

The likelihood a language model will generate factually incorrect statements about your brand in AI answers — wrong product names, incorrect pricing, false claims about features or team, outdated information. Higher for brands with sparse, inconsistent or low-authority web presence. Mitigated by entity grounding, schema markup, consistent entity naming, and high-quality primary-source coverage across the web.

A proposed standard file served at /llms.txt that provides AI systems with a curated, machine-readable summary of a site's most important pages. Proposed September 2024; by May 2026 it has ~10% adoption across 300,000 domains (SE Ranking). Empirical evidence of citation uplift is absent — ALLMO's analysis of 94,000+ cited URLs found no measurable uplift. Google has explicitly confirmed it does not use llms.txt. Legitimate as a developer-tooling interface (IDE agents fetch it); not yet a GEO signal. Source: Limy analysis, May 2026.

How many of the buyer's realistic questions about your category trigger an AI citation for your brand. Measured across three intent bands: discovery ("best tools for X"), comparison ("A vs B"), and validation ("is A reliable / pricing"). Full prompt coverage requires different content assets for each band. Comparison queries trigger AI Overviews 95% of the time (Seer 2026) — the highest-priority band for most B2B brands.

The complete set of queries a buyer might realistically ask an AI engine about your product category. Defining the prompt space before sampling is the methodological foundation of share-of-model measurement. A defensible prompt space is weighted by intent band frequency, reviewed by human editors, and versioned for historical comparison.

The dominant architecture of AI answer engines in 2026. A retrieval layer fetches candidate documents from an index; a language model generates an answer using those documents as context; citations in the answer refer to the retrieved documents. GEO optimises for two RAG stages: (1) being in the candidate set (crawlability, freshness, authority) and (2) being preferred by the ranking layer (specificity, structure, sourcing).

Structured data (typically JSON-LD) embedded in HTML that helps AI retrieval systems classify and extract content. Schema markup improves LLM discoverability by 67% (Yext, 2026). High-priority types for GEO: Organization (entity grounding), Article (content context + authorship), FAQ (direct answer extraction), HowTo, BreadcrumbList. Implemented in `<script type="application/ld+json">` in the `<head>`.

A 4-point scoring dimension in AI visibility measurement: Accurate-positive (correct facts, favourable framing), Accurate-neutral, Inaccurate (factual errors / hallucination), Absent. Sentiment quality matters because being mentioned inaccurately is often worse than not being mentioned — a hallucinated product claim can undermine buyer trust at the exact moment of highest intent.

The primary KPI of a GEO programme. Measures how often, across a representative sample of category-relevant prompts and engines, a brand is mentioned. A defensible share-of-model figure must disclose: prompt set and weighting, engine set and model versions, sample size, date window, and sentiment rubric. Without these five disclosures, the figure is marketing, not measurement. Share-of-model compounds like domain authority — brands cited today are cited more in the future.

Citation share measured specifically against a defined competitive set. Different from share-of-model in that SoV is inherently relative: if ChatGPT mentions your brand in 40% of category prompts but mentions your top competitor in 72%, your SoV is low despite reasonable absolute presence. Tracking SoV prevents teams from confusing absolute improvement with competitive advantage.

A query resolved inside the AI interface without the user visiting any external website. AI answer engines dramatically accelerate zero-click resolution. Data: organic CTR on AI Overview-present queries collapsed from 1.76% to 0.61% between January and September 2025, before recovering to 2.4% by February 2026 (Seer Interactive, 2.43B impressions). The implication: brand visibility in AI answers must generate trust and recall even when no click occurs.